Learning human arm movements by imitation evaluation of a biologically-inspired connectioni
Learning human arm movements by imitation:Evaluation of a
biologically-inspired connectionist architecture
Aude Billard&Maja Matari′c
Computer Science Department,University of Southern California,
SAL230,Los Angeles90089,U.S.A
Tel:+1-213-740-64-16,Fax:+1-213-740-75-12,
billard,mataric@https://www.wendangku.net/doc/051602861.html,
Abstract.This paper is concerned with the evaluation of a model of human imitation of arm movements.The
model consists of a hierarchy of arti?cial neural networks,which are abstractions of brain regions involved in
visuo-motor control.These are the spinal cord,the primary and pre-motor cortexes(M1&PM),the cerebellum,
and the temporal cortex.A biomechanical simulation is developed which models the muscles and the complete
dynamics of a37degree of freedom humanoid.Input to the model are data from human arm movements
recorded using video and marker-based tracking systems.
The model’s performance is evaluated for reproducing reaching movements and oscillatory movements of the
two arms.Results show a high qualitative and quantitative agreement with human data.In particular,the model
reproduces the well known features of reaching movements in humans,namely the bell-shaped curves for the
velocity and quasi-linear hand trajectories.Finally,the model’s performance is compared to that of humans
performing the same imitation task.It is shown that the model’s reproduction is better or comparable to that of
humans.
1Introduction
The goal of robotics is to have robots become a part of human everyday lives,in the roles of caretakers for the elderly and disabled,assistants in surgery and rehabilitation,and machine pets and toys for children.For this goal to be reached,robots will need to be programmed and interacted with in ways that are more natural and ef?cient than their current counterparts.A key challenges to making this possible is developing?exible motor skills in order to give robots the ability to be programmed and interacted with more easily and naturally,and to assist humans in various tasks.A very challenging and exciting area of current research is concerned with developing human-like robots(humanoids)for assisting humans in medical surgery[28,30]and rehabilitation[3],for providing help in everyday tasks to the elderly and the disabled[48],and for replacing humans in low-level industrial tasks and unsafe areas[20,24](including space,nuclear,and waste management industries).
To provide robots with human-like capabilities,and in particular with sophisticated motor skills for?exible and precise motions,is a very dif?cult task,requiring important low-level programming(with high cost)for?ne tuning of the motor parameters and re-calibration of sensor processing[12,39].An alternative is to provide the robot with learning or adaptive capabilities,which can be used for on-and/or off-line optimization of prede?ned motor control parameters[8,22,45].Particularly challenging is the problem of how to teach a robot new motor skills,without going through reprogramming,but instead through demonstration.In such a scenario,the robot learns novel motor sequences by replicating those demonstrated by a human instructor and by tuning its motor program descriptions so as to successfully achieve the task.The method is interesting because it allows the robot to be programmed and interacted with merely by human demonstration,a much more natural and simple means of human-machine interface;and it makes the robot?exible with respect to the tasks it can be taught and,thus, facilitates the end-use of robotic systems.
The?rst robotics work to address imitation was focused on assembly task-learning from observation.Typically, a series of visual images of a human performing a simple object moving/stacking tasks was recorded,segmented, interpreted,and then repeated by an industrial non human-like robotic arm[19,18,26,16,21].These efforts consti-tute a signi?cant body of research in robotics,and contribute to video segmentation and understanding.However, they provide highly task-speci?c solutions,with little?exibility for applying the same algorithm to imitation after types of movements and tasks.More recent efforts,including our own,have been oriented toward analyzing the underlying mechanisms of imitation in natural systems and modeling those on arti?cial ones[4,6,5,9,10,31,42, 41].The endeavor,there,is,on the one hand,to build biologically plausible models of animal’s imitative abilities, and,on the other hand,to develop architecture for visuo-motor control and learning in robots which would show some of the?exibility of natural systems.
Our work wishes to complement other above approaches,by investigating a connectionist based model,cou-pled to a complete biomechanical simulation of a humanoid.We follow neuroscience studies of primate motion recognition and motor control.Speci?cally,our work is driven by the observation that1)visual recognition of movements is done in both excentric and egocentric frame of reference[35,46];2)that a neural system,the mirror neuron system,encapsulates a high-level representation of movements,the link between visual and motor repre-sentation[38,11];and3)that motor control and learning is hierarchical and modulates(evolutionary)primitive motor programs(central pattern generators,located in primates’spinal cord[44]).
Our model is composed of a hierarchy of arti?cial neural networks and gives an abstract and high-level rep-resentation of the neurological structure underlying primates brain’s visuo-motor pathways.These are the spinal cord,the primary and pre-motor cortexes(M1&PM),the cerebellum and the temporal cortex.The model has?rst been evaluated in a pair of demonstrator-imitator humanoid avatars with65degrees of freedom[6]for learning by imitation gestures and complex movements involving all the avatar’s limbs.In this paper,we evaluate the model’s performance at reproducing human arm movements.A biomechanical simulation is developed which models the muscles and the complete dynamics of a37degree of freedom humanoid1.The aim of these experiments is to evaluate the realism of the model and the dynamic simulation at modeling human imitation.
The rest of the paper is organized as follows.In Section2,we describe in detail the model,and,in particular, the visual processing of the data and the learning algorithm.In Section3,we evaluate the model’s performance on a series of experiments for reproducing human arm motion,namely reaching movements and oscillatory movements of the two arms.We compare the model’s performance to that of humans in the same imitation task.Section4 concludes this paper with a short summary of the presented work.
2The model
We have developed a highly simpli?ed model of primate imitative ability[6](see Figure1).This model is biologically inspired in its function,as its composite modules have functionalities similar to that of speci?c brain regions,and in its structure,as the modules are composed of arti?cial neural architectures(see Figure2).It is loosely based on neurological?ndings in primates and incorporates abstract models of some brain areas involved in visuo-motor control,namely the temporal cortex(TC),the spinal cord,the primary motor cortex(M1),the premotor area(PM)and the cerebellum.
2.1Brief description of the modules
Visual information is processed in TC for recognition of the direction and orientation of movement of the demon-strator’s limbs relative to a frame of reference located on the demonstrator’s body.That is,the TC module takes as input the Cartesian coordinates of each joint of the demonstrator’s limbs in an excentric frame of reference(whose origin is?xed relative to the visual tracking system).It then transforms these coordinates to a new set of coordi-nates relative to an egocentric frame of reference.Our assumption of the existence of orientation-sensitive cells in an egocentric frame of reference in TC is supported by neurological evidence in monkeys[35,36]and humans [2,25,46].The vision system also incorporates a simpli?ed attentional mechanism which is triggered whenever a signi?cant change of position(relative to the position at the previous time step)in one of the limbs is observed. At this stage of the modeling and given the simplicity of this module,the attentional module does not relate to any speci?c brain area.The attentional mechanism creates an inhibition,preventing information?ow from M1 to PM and further to the cerebellum,therefore allowing learning of new movements only when a change in the limb position is observed.In Section2.2,we describe the motion tracking system we used in the experiments and explain in more detail the stages of visual processing in the TC module.
Motor control in our model is hierarchical with,at the lowest level,the spinal cord module,composed of primary neural circuits(central pattern generators(CPGs)[44]),made of motor neurons and interneurons2(see Section2.3).The motor neurons in our simulation activate the muscles of the humanoid avatar,see Section2.5.The M1module monitors the activation of the spinal networks.Nodes in M1are distributed following a topographic map of the body.
Learning System
Fig.1.The architecture consists of seven modules which give an abstract and high-level representation of corresponding brain areas involved in visuo-motor processing.The seven modules are:the attentional and temporal cortex modules,the primary motor cortex and spinal cord modules,the premotor cortex and cerebellum module,and the decision module.
Learning of movements is done in the PM and cerebellum modules.These modules are implemented using the Dynamical Recurrent Associative Memory Architecture(DRAMA)[7]which allows learning of time series and of spatio-temporal invariance in multi-modal inputs(see Section2.4for details).Finally,the decision module controls the transition between observing and reproducing the motor sequences,i.e.it inhibits PM neural activity due to TC (visual)input to?ow downwards to M1(for motor activation).It is implemented as a set of if-then rules and has no direct biological inspiration.
Neurons in the PM module respond to both visual information(from TC)and to corresponding motor com-mands produced by the cerebellum.As such,they give an abstract representation of mirror neurons.Mirror neurons refer to neurons located in the rostral part of inferior premotor area6in monkey[11,38],which have been shown to?re both when the monkey grasps an object and when it observes another monkey or a human performing a similar grasp.
In the next section,we describe in more details the visual,motor,and learning parts of our model.
2.2Visual Segmentation
Data for our experiments(see Section3)are recordings of human motion.The?rst set of data was recorded using a motion-tracking system.The system we used is capable of selecting a collection of features from the moving image,based on a constrained(un-occluded and unambiguous)initial position and kinematic model of a generic adult human(See[47]for a detailed description).Tracking is done off-line and based on image frequency of15 Hz.The system allows tracking of the upper body in the vertical plane,where the body features correspond to those of a stick?gure(see Figure3).It calculates the positions(relative to a?xed,excentric frame of reference) of nine points on the body:two located on the wrists,two on the elbows,two on the shoulders,one on the lower torso,one on the neck and one on the head.
A second set of human arm data,used in the experiments,was gathered by Matari′c and Pomplun in a joint interdisciplinary project conducted at the National Institutes of Health Resource for the Study of Neural Models of Behavior,at the University of Rochester[32,37].Subjects watched and imitated short videos of arm movements,
TC
Fig.2.Schematics of the neural structure of each module and their interconnections.
Fig.3.(left)Motion tracking system of human movement.(Right)The Cosimir simulator.
while wearing FastTrak marker mechanism for recording the positions of4markers on the arm:at the upper arm, near the elbow,the wrist,and the hand.
In the experiments,these Cartesian coordinates are input to the temporal cortex module(TC)of our model, in which they are processed in four stages.Data are?rst transferred into a frame of reference relative to the demonstrator’s body,by calculating the joint angles of the elbows and shoulders.In a second stage,a low-pass ?lter is applied to the calculation of the angular velocity for each of the four joints.This stage corresponds to the attentional mechanism of Figure1.This allows us to eliminate small arm movements which we consider noise for these experiments.These small motions are due to two factors:1)the locations of the nine points of reference of the tracking are imprecise;the coordinates are extrapolated across three time steps of recording;2)because of the interaction torques across the body,movement of one limb results in small motions of the rest of the body.These small movements are noise to us,as we wish to recognize only voluntary movements(as opposed to movements made to compensate for the interaction torques).Since shoulders and elbows have different dynamics,due to their different lengths and muscular composition,we applied different?lter parameters are applied to each.The?ltering process depends on a set of2parameters per DOF.These2are thresholds de?ning1)the minimum displacement (in joint angle)for detecting a motion,2)the minimum time delay during which no displacement greater than has been observed.The latter is then considered as a stop of the motion or small,noisy movements.Table1 shows the values we used for the experiments reported in Section3.Note that in the experiments,we used at most2 (abduction and?exion)of the3DOFs of the shoulders,as the third DOF,humeral rotation,was not recorded by any of the two tracking systems.Figure6show the results of the visual segmentation for three oscillatory movements of the two arms.Only the large movements are segmented.
Table1.Thresholds(in degrees)for visual?ltering.LSx is the DOF x of the left shoulder.LE is the left elbow.(in radians)is the minimum displacement for detecting a motion.(in recording cycles)is the minimum time delay during which no displacement greater than has been observed.
Experiment
PI/16
LSy15
PI/16
RSy15
PI/8
RE10
In a third stage,we calculate the direction of movement of each limb relative to the limb to which it is attached (elbow relative to shoulder and shoulder relative to the torso).The direction of movement is positive or negative depending on whether the limb moves upwards or downwards,respectively.In the fourth stage,the TC module activates a series of cells coding for the possible joint angle distributions.There are two cells per degree of freedom (DOF)per joint,coding for positive and negative direction of movement,respectively.The output of the cells encodes both the direction and speed of the movement.The faster the speed,the greater the output excitation of the cell.Only one cell of the pair is active at a time.If both cells are inactive,the limb is not moving.The decomposition of the limbs’motion can easily be mapped to the muscular structure of the imitator;each DOF of a limb is directed by a pair of?exor-extensor muscles.Upward and downward directions of movement correspond to the activation of the extensor and?exor muscles,respectively.
In summary,the visual module performs four levels of processing on the data:1)a transformation from extrinsic to intrinsic frame of reference,2)?ltering of small and noisy motions,3)a parameterization of the movements in terms of speed and direction,and4)segmentation of the motion,based on changes in velocity and movement direction.
2.3Motor control
Spinal Cord Module
In our model,motor control is hierarchical.On the lowest level of motor control is the spinal cord module. It is composed of primary neural circuits made of motor neurons(afferent to the muscles and responsible for the muscle activation or inhibition)and interneurons.
In our experiments,the spinal circuits are built-in and encode extending and retracting arm movements,as well as rhythmic movements of legs and arms involved in the locomotion,following a biological model of the walking neural circuits in vertebrates[17].The neurons of the spinal cord module are modeled as leaky-integrators,which compute the average?ring frequency[15].According to this model,the mean membrane potential of a neuron is governed by the equation:
(1)
where represents the neuron’s short-term average?ring frequency,is the neuron’s bias,is a time constant associated with the passive properties of the neuron’s membrane,and is the synaptic weight of a connection from neuron to neuron.
Motor Cortex Module:M1
The primary motor cortex(M1)module contains a motor map of the body(similar to the corresponding brain area[34]).It is divided into layers of three neuron networks,each activating distinct(extensor-?exor)muscle pairs(see Figure2).The three-neuron network allows for independently regulating the amplitude(two nodes, one for each muscle)and the frequency(one node)of the oscillation of the corresponding?exor-extensor pair, similar to[17].An oscillation of a limb segment is generated by activating all three neurons,allowing a small time delay between activation of the?rst and second neuron,thus creating an asymmetry between the two motor neurons’activity and the corresponding muscle contraction.Motion of a single muscle(?exor or extensor)is obtained by activating only one of the two amplitude nodes,while keeping the frequency node at zero.The speed of the movement,i.e.,the speed of contraction of the muscle,is controlled by increasing the output value of the amplitude neuron and consequently that of the corresponding motor neuron in the spinal cord.The amplitude of the movement(in the case of one-muscle activation)is controlled by the duration of the neuron activation.The longer the activation of the amplitude neuron(and subsequently of the motor neuron),the longer the duration of muscle contraction,the larger the movement.
M1receives sensory feedback,in the form of joint angle position,from the spinal cord module.Each motor area of M1receives sensory feedback from its related sensory area (arm area receives feedback on joint positions of the shoulder joints).This is used to modulates the amplitude or speed of the movement,by increasing or decreasing (for smaller or larger speed)the output of the M1nodes.The sensory feedback provides inhibition;the larger the feedback,the slower the movement.In the experiments of Section 3.1,this is used to modulate reaching movements.When the movement starts,the sensory feedback is at its minimum and consequently the tonic input (i.e.,the amplitude of the M1nodes’output)is at its maximum.When the arm has reached half of the required distance,the sensory feedback is at its maximum and,consequently,the tonic input is decreased to 10%of its maximum.The arm stops shortly afterwards when the torque produced by the muscle (proportional to the motor neuron’s output,see Section 2.5)equates that of gravity.
Premotor Cortex Module:PM
The PM module creates a direct mapping between the parameterization of the observed movement in TC,following visual segmentation,and that used for motor control in M1.In TC,the observed motion is segmented in terms of speed,direction and duration of movement (the delay between two changes in velocity and motion direction)of each limb (see Section 2.2).In M1,speed and direction of movement of each limb CPG (in the spinal cord)are controlled by the amplitude of the nodes which project to the relevant interneurons.PM nodes transfer the activity of the TC nodes (observation of a speci?c movement)into an activity pattern of M1nodes (motor command for the corresponding movement).A large output activity in TC cells (comprised between 0and 1)will lead to an important output from PC nodes,and further from M1nodes which further to activation of the corresponding amplitude node.Duration of movement is proportional to the duration of activation of the amplitude node.Learning of the movements consists,then,of storing the sequential activation (recording the amplitude and the time delay)of each of the TC nodes,and mapping these to the corresponding M1nodes.This will be further explained in Section 2.4.
Decision module
Finally,the execution of a movement (as during rehearsal of the motion in the experiments,see Section 3)is started by the decision module,by activating one of the cerebellum nodes (the node which encodes the corre-sponding sequence of muscle activation,described in 2.4).The activity of the cerebellum node is passed down to the nodes of the premotor cortex,which encode co-activation of the muscle in a speci?c step of the sequence (de-scribed in Section 2.4),and,further,down to the nodes of the second layer of primary motor cortex (M1).Finally,the activity of the nodes in the second layer of M1activates the nodes in the spinal cord module,which further activates the motor neurons and these the simulated muscles of the avatar.2.4The learning modules
τ0
t j i
0t +Fig.4.Schematics of one connection from unit i and unit j.Each connection of the DRAMA network is associated with two parameters,a weight
and a time
parameter
.Weights correspond to the synaptic strength,while the time parameter speci?es a synaptic delay.Each unit has a self connection.Retrieval follows
a winner-take-all rule on the weights.
Learning of motor sequences is done by updating the connectivity between the primary cortex (M1),the pre-motor cortex (PM),and the cerebellum modules.PM and cerebellum modules consist of a Dynamical Recurrent
Associative Memory Architecture (DRAMA)[7],a fully recurrent neural network without hidden units.Similarly to time delay networks [29],each connection is associated with two parameters,a weight and a time parameter
(see Figure4).Weights correspond to the synaptic strength,while he time parameter speci?es a synaptic delay, that is a delay on the time required to propagate the activity from one neuron to the other.Both parameters are modulated by the learning in order to represent the spatial()and temporal()regularity of the input to a node. The parameters are updated following Hebbian rules,given by Equations2and3.Learning starts with all weights and time parameters set to zero,unless speci?ed differently to represent prede?ned connection(as between PM-M1 modules,see Section2.3).
(2)
(3)
where is a constant factor by which the weights are incremented.
In the present experiment,learning across TC-PM,PM-M1and PM-cerebellum consists of building up the connectivity of nodes across these modules so as to represent spatio-temporal patterns of activation in the TC and PM modules,respectively.The connectivity PM-M1is constructed simultaneously to that of TC-PM to represent the isomorphism between visual and motor representation.
In DRAMA,the neuron activation function follows a linear?rst order differential equation given by Equation 4,below.
(4)
is the identity function for input values less than1and saturates to1for input values greater than1(if and otherwise)and is the retrieving function whose equation is given in5.
(5)
The function is a threshold function that outputs1when and0otherwise.The factor is a error margin on the time parameter.It is equal to in the simulations,allowing a10%imprecision in the record of the time delay of units co-activation.The term is a threshold on the weight.It is equal to
where and are the mass and the moment of inertia of link.is the position vector of joint compared to the center of mass of link.
These dynamics equations are solved using MathEngine’s Fastdynamics3which computes the internal forces keeping the links connected,as well as the forces due to contacts,while the external forces such as the muscles torques,the forces due to gravity and to the damping due to the air are given by the user.
Muscle torques A muscle is simulated as a combination of spring and damper[27].The torque exerted on each joint is determined by a pair of opposed?exor and extensor muscles.These muscles can be contracted by input signals from motor neurons,which increase their spring constant,and therefore reduce their resting length.The torque acting at a particular joint is therefore determined by the motoneuron activities(and)of the opposed ?exor and extensor muscles:
(8)
where is the difference between the actual angle of the joint and the default angle.The different coef?cients,,,and determine,respectively,the gain, the stiffness gain,the tonic stiffness,and the damping coef?cient of the muscles.
3Experiments
We present a series of experiments in which we measured the performance of the model at reproducing well-known features of human arm movement during reaching and the precision with which the model reproduced sequences of oscillatory arm movements.We also compared the model performance to humans performance imitating the same arm movements.
The model was implemented on eight sets of human arm motions.The?rst three?rst sets were recorded using the video tracking system described in[47],and consisted of2D oscillatory movements of the two arms in the vertical plane(lifting up and down the shoulders and bending the elbows).The other?ve sets were recorded using a FastTrak marker-based system(see[37]for a complete report)and consisted of3D oscillatory movements of the left arm.
3.1Reaching movements
We evaluated the model’s performance in reproducing reaching movement of the left arm using the data recorded using the FastTrak system(see Section2.2).In this experiment,the model was given the target of the trajectory (i.e.the desired angle for each DOF of the shoulder and elbow)as input for the reproduction.These values were used by the spinal cord module of the model to modulate the value of the sensory feedback.There is no learning in this example.The model’s prede?ned connectivity for reaching(in the PC module)is exploited to generate the motions.We tested the correctedness of the model in reproducing two main features associated with human arm movements,namely the bell-shape velocity curve and the quasi-straight curvature of the hand trajectory in space [1,33,43].
Figure5(bottom)shows the trajectory,velocity pro?le,and the hand path of the avatar’s hand during a reaching movement directed towards a point at25degrees in the x direction and30degrees in the z direction.Figure5(top) shows the same values for the trajectory of the human hand in a similar reach(aimed at the same target).In both avatar and human movements,the velocity pro?les for the biggest directions of movements(x and z)follow a bell-shape curve.In the direction of small movements(y axis),which result from internal torques caused by movement in the two other degrees of freedom,the velocity pro?le is made of small oscillatory movements in both avatar and human.Similarly to the human data,the avatar’s hand trajectory is smooth,reaching its sharpest slope at middle distance(a fact re?ected by the bell-shape velocity pro?le).In our model,the slow increase of velocity for the ?rst half of the distance is due to the smooth increase of neural activation of the motor neuron(the motor neuron’s output is directly proportional to the elasticity constraint of the modeled muscles,see Equation8),which follows a sigmoid(see Equation1).The plateau and decrease of the velocity starting at midistance is due to1)the damping factor in Equation8,a muscle property,and to2)a property of the controller,which decreases the tonic input(from PM and M1nodes)sent to the motor neurons when receiving sensory feedback(relative position in joint angles) from the spinal cord module indicating that about half of the requested distance had been achieved.
Fig.5.Human data(top3graphs),avatar data(bottom3graphs).In each3graph sets:trajectory(top),velocity pro?le(middle)and path(bottom)of the hand in x,y,z directions during a reaching movement directed towards a point at25degrees in the x direction and30degrees in the z direction.
3.2Oscillatory arm movements
This section describes results using the three motion sets recorded with the video tracking system,which consisted of lifting up and lowering left and right upper arms(vertical rotation around the shoulders),while bending and extending the lower arms(rotation around the elbows),respectively.For each set,the motion was repeated twice.
For these experiments,the reproduction of the movement was not driven by a target in joint angle as in the previous Section.Here,observed motion of each limb were fed continuously to the TC module.Each change of movement would trigger the TC cells.Their activity,which encoded the new orientation and speed of the movement,would be passed further to the PC and cerebellum module to learn the sequence of movement.At the end of the observation,the cerebellum and PC are activated by the decision module to trigger rehearsal of the learned sequence.
Figure6shows superimposed trajectories of the left and right shoulders and elbows of the avatar and the human for the three sets of motions.The black vertical lines show the instants during the movement at which the visual segmentation triggered(detecting a start or end of the motion based on velocity and direction changes).The avatar’s reproduction shows a qualitative and quantitative agreement with the human movement.It reproduces all the large movements of the shoulders and elbows,with a similar amplitude.A good reproduction of the amplitude of the movement is obtained in the model by keeping a good measure of the speed of the observed movement.The speed of the movement is transmitted by the amplitude of the output of the TC cells(see Section2.2),which is then recorded in the PM weights and further transmitted to motor neurons(in the spinal cord)as the amplitude of PM and M1nodes’output.In the above example,we chose a1%precision in the speed recording.By varying this precision,one can approximate the precision with which human can make similar measurement.We discuss this in the next section.
3.3Comparison with human imitative performance
Using the data gathered in[37]on human imitation of arm movement,we evaluated the precision within which humans reproduce arm movements.Figure7shows the trajectories of the left hand of each of four human imitators, that of the human demonstrator and that of the avatar’s reproduction of the same trajectory.
Fig.6.Superimposed trajectories of left/right shoulder/elbow of the avatar and the human during the three movement sequences(from top to bottom).
Fig.7.Trajectories of hand motion of four human subjects and the avatar imitating an oscillatory movement of the left arm,demonstrated by another human subject.First top row:Human demonstration;rows2-5,imitation by four human subjects;6th row,imitation by the humanoid avatar.
The imitation by the human subjects is qualitatively similar to the demonstration,as they correctly reproduced the two oscillations in the z direction.However,some subjects produced movements in the x and y directions as well.The amplitude and timing of the movement is not reproduced very well.In these two respects,the avatar’s reproduction is as good as that of the human.We measure the precision of the imitation following two criteria: 1)the qualitative similarity between demonstrator’s and imitator’s limbs’trajectories(hand path in extrinsic co-ordinates for reaching and shoulder,elbow joint angles for other movements),obtained by comparing the number of maxima and minima of each curve;2)the quantitative similarity of the trajectories in terms of amplitude and speed.We measure(Max(Imitator)/Max(Demonstration)),the ratio between maxima of amplitude,and, (=—t(max(Imitator)-t(max(demonstrator)—/T)the ratio of the time difference between two maxima over the du-ration.This is a straightforward measure of the observable dissimilarities between the two trajectories,which we will use,in future experiments,as feedback to tune the learning so as to improve the reproduction.is a direct measure of the amplitude difference between the movements,while is an indirect measure of the speed differ-ence between the movements.In[37],we present other measures of similarity between the trajectories in the same reaching tasks as presented here and discuss the performance of each.
Table2.Qualitative comparisons of human and avatar imitative performance.is the ratio between maxima of amplitude and is the ratio of the time difference between two maxima over the duration of the demonstration for human and avatar trajectories.Data are mean values and standard deviation across imitation of eight data sets.
Human
Table2shows the mean values of these measures across imitation of the eight data sets for human imitation and avatar replication.Avatar and human performance following these measures are quantitatively similar.Both show an imprecision of over20percent on average for reproducing the amplitude and the speed of the movement.
This similarity between human and avatar data is encouraging,as the long term goal of this study is to de-sign a model of human ability to learn movements by imitation.Further work will focus on developing precise measures of trajectories similarities and on determining the in?uence of each parameter of the model and of the biomechanical simulation on the model’s performance.
4Conclusion
This paper presented a series of experiments to evaluate the performance of a connectionist model for imitating human arm movements.The model is composed of a hierarchy of arti?cial neural network models,which each give an abstract representation of the functionality of some brain area involved in motor control.These are the spinal cord,the primary and pre-motor cortexes(M1&PM),the cerebellum,and the temporal cortex.
The model was implemented in a biomechanical simulation of a humanoid avatar with37degrees of freedom. Data for the imitation were recordings of human arm motions for reaching and oscillatory movements.To validate the model using real data,as opposed to simulated ones,and using a complete biomechanical simulation was very important to us,as our goal is to implement the system on a real robotic platform.
Results showed that the model could reliably reproduce all motions,while data were highly noisy.We measured a good quantitative agreement between simulated and real data,based on an error measure on the amplitude and speed of the movement.Moreover,the measured error in the model’s reproduction was comprised within the range of error made by humans engaged in the same imitation task.These results suggest that the connectionist model, coupled to the biomechanical simulation,could be a good?rst approximation of human imitation.Future work will aim at evaluating further the model’s performance on more data and at comparing its performance to other models of human motor control,such as[23,14,13,42].
Acknowledgments
Warmest thanks to Auke Ijspeert for his precious comments all along this project.Many thanks to Stefan Weber for providing the data and the vision based motion tracking software.Many thanks to the Robotics Institute at the University of Dortmund for providing the Cosimir simulator.This work was supported in part by the National Science Foundation,CAREER Award IRI-9624237to M.Matari′c and in part by the Of?ce of Naval Research.Aude Billard was supported partly by a fellowship from the Medicus Foundation,New York.
References
1.W Abend,E Bizzi,and P Morasso.Human arm trajectory formation.Brain,105:331–348,1981.
2.R.A.Andersen,https://www.wendangku.net/doc/051602861.html,wrence,D.C.Bradley,and J.Xing.Multimodal representation of space in the posterior parietal
cortex and its use in planning movements.Annualr review of neuroscience,20:303–330,1997.
3. A.K.Bejczy.Towards development of robotic aid for rehabilitation of locomotion-impaired subjects.Proceedings of the
First Workshop on Robot Motion and Control,pages9–16,1999.
4.L.Berthouze,P.Bakker,and Y Kuniyoshi.Learning of oculo-motor control:a prelude to robotic imitation.In Proceedings
of the1996lEEE/RSJ International Conference on Intelligent Robots and Systems’96,pages376–381,1996.
5. A.Billard.Imitation:a means to enhance learning of a synthetic proto-language in an autonomous robot.In C.Nehaniv
and K.Dautenhahn,editors,Imitation in Animals and Artifacs.MIT Press,2000.To appear.
6. A.Billard.Learning motor skills by imitation:a biologically inspired robotic model.Cybernetics&Systems Journal,
special issue on Imitation in animals and artifacts,2000.To appear.
7. A.Billard and G.Hayes.Drama,a connectionist architecture for control and learning in autonomous robots.Adaptive
Behavior,Vol.7:1,pages35–64,1999.
8.V.R.de Angulo and C.Torras.Self-calibration of a space robot.IEEE Transactions on Neural Networks,8:4:951–963,
1997.
9.J.Demiris.Movement imitation mechanisms in robots and humans.PhD thesis,Dept.of Arti?cial Intelligence,University
of Edinburgh,May1999.
10.J.Demiris,S.Rougeaux,G.M.Hayes,L.Berthouze,and Y.Kuniyoshi.Deferred imitation of human head movements by
an active stereo vision head.In Proceedings of the6th IEEE International Workshop on Robot Human Communication, pages88–93.IEEE Press,Sendai,Japan,Sept.1997.
11.G.di Pellegrino,L.Fadiga,L.Fogassi,V.Gallese,and G.Rizzolati.Understanding motor events:a neurophysiological
study.Experimental Brain Research,91:176–180,1992.
12.J.Dias,A.de Almeida,H.Araujo,and J.Batista.Camera recalibration with hand-eye robotic system.IEEE International
Conference on Industrial Electronics,Control and Instrumentation,1:1923–1928,1991.
13.Hiroaki Gomi and Mitsuo Kawato.Equilibrium-point control hypothesis examined by measured arm stiffness during
multijoint movement.science,272:117–120,1996.
14.S.R.Goodman and G.L.Gottlieb.Analysis of kinematic invariances of multijoint reaching movement.Biological
Cybernetics,73:4:311–322,1995.
15.J.J Hop?eld.Neurons with graded response properties have collective computational properties like those of two-state
neurons.In Proceedings of the National Academy of Sciences,volume81,pages3088–3092.Washington:The Academy, 1984.
16.Geir E.Hovland,Pavan Sikka,and Brenan J.McCarragher.Skill acquisition from human demonstration using a hidden
markov model.In Proceedings,IEEE International Conference on Robotics and Automation,pages2706–2711,Min-neapolis,MN,1996.
17. A.J.Ijspeert,J.Hallam,and D.Willshaw.Evolving swimming controllers for a simulated lamprey with inspiration from
neurobiology.Adaptive Behavior,7(2):151–172,1999.
18.Katsushi Ikeuchi,Masato Kawade,and Takashi Suehiro.Assembly task recognition with planar,curved,and mechanical
contacts.In Proceedings of IEEE International Conference on Robotics and Automation,Atlanta,GA,1993.
19.Katsushi Ikeuchi,Takashi Suehiro,Peter Tanguy,and Mark Wheeler.Assembly plan from observation.Technical report,
Carnegie Mellon Univeristy Robotics Institute Annual Research Review,1990.
20.N.Ishikawa and K.Suzuki.Development of a human and robot collaborative system for inspecting patrol of nuclear power
plants.6th IEEE International Workshop on Robot and Human Communication,pages118–123,1997.
21.Michael Kaiser.Transfer of elementary skills via human-robot interaction.Adaptive Behavior,5(3–4),1997.
22. F.Kanehiro,M.Inaba,and H.Inoue.Action acquisition framework for humanoid robots based on kinematics and dynamics
adaptation.In IEEE International Conference on Robotics and Automation,volume2,pages1038–1043,1999.
23. A.Karniel and G.F.Inbar.A model for learning human reaching movements.Biological Cybernetics,77,Issue3:173–183,
1997.
24.K.Kawamura,D.M.Wilkes,T.Pack,M.Bishay,and J.Barile.Humanoids:Future robots for home and factory.In
Proceedings of the First International Symposium on Humanoid Robots,Waseda University,Tokyo,Japan,pages53–62, 1996.
25. C.Kertzman,U.Schwarz,T.A.Zef?ro,and Mark Hallett.The role of posterior parietal cortex in visually guided reaching
movements in humans.Experimental Brain Research,114,Issue1:170–183,May131997.
26.M.I.Kuniyoshi and I.Inoue.Learning by watching:Extracting reusable task knowledge from visual observation of human
performance.IEEE Transactions on Robotics and Automation,vol.10,no.6,pages799–822,1994.
27.F Lacquaniti and J.F Soechting.Simulation studies on the control of posture and movement in a multi-jointed limb.
Biological Cybernetics,54:367–378,1986.
28.M.A.Lewis and G.A.Bekey.Automation and robotics in neurosurgery:Prospects and problems.Chapter6in Neuro-
surgery for the Third Millineum,pages65–79,1992.
29. D.T.Lin,P.A.Ligomenides,and J.E.Dayhoff.Learning spatio-temporal topology using an adaptive time-delay neural
network.In Proceedings of World congress on neural networks,Portland,OR,volume1,pages291–294,1993.
30.Y.Louhisalmi and T.Leinonen.On research of directly programmable surgical robot.Engineering in Medicine and Biology
Society,Bridging Disciplines for Biomedicine.,18th Annual International Conference of the IEEE,1:229–230,1997. 31.M.J Matari′c.Sensory-motor primitives as a basis for imitation:Linking perception to action and biology to robotics.In
Chrystopher Nehaniv and Kerstin Dautenhahn,editors,Imitation in Animals and Artifacts.The MIT Press,2000.
32.M.J.Matari′c and Marc Pomplun.Fixation behavior in observation and imitation of human movement.Cognitive Brain
Research,7(2):191–202,1998.
33.P Morasso.Spatial control of arm movements.Experimental Brain Research,42:223–428,1981.
34.W.Pen?eld and T.Rassmussen.The Cerebral Cortex of Man:A clinical Study of Localisation of Function.New York:
Macmullan,1950.
35. D.I.Perret,M Harries,R.Bevan,S.Thomas,P.J.Benson,A.J.Mistlin,A.J.Chitty,J.K.Hietanene,and J.E.Ortega.
Frameworks of analysis for the neural representation of animate objects and actions.Journal of Experimental Biology, 146:87–113,1989.
36. D.I.Perret,M Harries,A.J.Mistlin,and A.J.Chitty.Three stages in the classi?cation of body movements by visual
neurons.In H.B et al.Barlow,editor,Images and Understanding,pages94–107.Cambridge University Press,1989. 37.Marc Pomplun and Maja J.Matari′c.Evaluation metrics and results of human arm movement imitation.In submitted to
umanoids-2000,MIT,Cambridge,MA,Sep7-8,2000,2000.
38.G.Rizzolati,L.Fadiga,V.Gallese,and L.Fogassi.Premotor cortex and the recognition of motor actions.Cognitive Brain
Research,3:131–141,1996.
39.J.Roning and A Korzun.A method for industrial robot calibration.IEEE International Conference on Robotics and
Automation,4:3184–3190,1997.
40.Freund.E.Rossmann,J.Projective virtual reality:Bridging the gap between virtual reality and robotics.IEEE Transaction
on Robotics and Automation;Special Section on Virtual Reality in Robotics and Automation,15:3:411–422,June1999.
www.irf.de/cosimir.eng/.
41.S.Schaal.Learning from demonstration.Advances in Neural Information Processing Systems,9:1040–1046,1997.
42.S.Schaal and D.Sternad.Programmable pattern generators.In Proceedings,3rd International Conferece on Computational
Intelligence in Neuroscience,pages48–51,Research Triangle Park,NC,1998.
43.L.E.Sergio and S.H.Scott.Hand and joint paths during reaching movements with and without vision.Biological
Cybernetics,122Issue2:157–164,1998.
44.P.S.G Stein,S.Grillner,A.I.Selverston,and D.G.Stuart.Neurons,Networks and Motor Behavior.A Bradford book:MIT
Press,1997.
45.S.Thrun.Explanation-Based Neural Network Learning-A Lifelong Learning Approach.Kluwer Academic,Boston,MA,
1996.
46.G.Vallar,E.Lobel,G.Galati,A.Berthoz,L.Pizzamiglio,and D.Le Bihan.A fronto-parietal system for computing the
egocentric spatial frame of references in humans.Experimental Brain Research,124:281–286,1999.
47.Stefan Weber.Simple human torso tracking from video.Technical Report IRIS-00-380,University of Southern California,
Institute for Robotics and Intelligent Systems,2000.
48.Song Won-Kyung,Lee He-Young,Kim Jong-Sung,Yoon Yong-San,and Bien Zeungnam.Kares:intelligent rehabilitation
robotic system for the disabled and the elderly.IEEE,20th Annual International Conference on Engineering in Medicine and Biology Society,5:2682–2685,1998.
统一身份认证-CAS配置实现
一、背景描述 随着信息化的迅猛发展，政府、企业、机构等不断增加基于Internet/Intranet 的业务系统，如各类网上申报系统，网上审批系统，OA 系统等。系统的业务性质，一般都要求实现用户管理、身份认证、授权等必不可少的安全措施，而新系统的涌现，在与已有系统的集成或融合上，特别是针对相同的用户群，会带来以下的问题： 1、每个系统都开发各自的身份认证系统，这将造成资源的浪费，消耗开 发成本，并延缓开发进度； 2、多个身份认证系统会增加系统的管理工作成本； 3、用户需要记忆多个帐户和口令，使用极为不便，同时由于用户口令遗 忘而导致的支持费用不断上涨； 4、无法实现统一认证和授权，多个身份认证系统使安全策略必须逐个在 不同的系统内进行设置，因而造成修改策略的进度可能跟不上策略的变化； 5、无法统一分析用户的应用行为 因此，对于拥有多个业务系统应用需求的政府、企业或机构等，需要配置一套统一的身份认证系统，以实现集中统一的身份认证，并减少信息化系统的成本。单点登录系统的目的就是为这样的应用系统提供集中统一的身份认证，实现“一点登录、多点漫游、即插即用、应用无关”的目标，方便用户使用。 二、CAS简介 CAS（Central Authentication Service），是耶鲁大学开发的单点登录系统（SSO，single sign-on），应用广泛，具有独立于平台的，易于理解，支持代理功能。CAS系统在各个大学如耶鲁大学、加州大学、剑桥大学、香港科技大学等得到应用。Spring Framework的Acegi安全系统支持CAS，并提供了易于使用的方案。Acegi安全系统，是一个用于Spring Framework的安全框架，能够和目前流行的Web容器无缝集成。它使用了Spring的方式提供了安全和认证安全服务，包括使用Bean Context，拦截器和面向接口的编程方式。因此，Acegi 安全系统能够轻松地适用于复杂的安全需求。Acegi安全系统在国内外得到了广
目标客户群体定位
关于“目标客户群体定位”的思考 关于“目标客户群体定位”的思考 众所周知，任何企业都是通过向产业链下游提供产品（服务）获取社会认同及股东收益的，我们统称这些购买企业产品的行为单元为客户。多数时候，企业无法将自己的产品功能丰富至可以服务于对同类产品有需求的所有客户的境界，无法在整个同业市场中实现价值传递。于是，企业针对自身的能力向特定的客户提供有特定内涵的产品价值，这些特定的客户就是“目标客户群体”。 为什么要寻找目标客户群体 随着我国经济市场化程度的不断加深及买方需求的多样化趋势，构成产业链的元素进一步分裂，市场细分成为了新世纪中国经济成熟的标志，为满足消费者日益细化的需求而衍生出许多细分行业使单元产业的价值链条愈见加长，通吃产业链的产品已经成为过去时，针对部分消费者（目标客户群体）的细分需求制定产品定位方可打造企业的核心竞争力。 目标客户群体的初步确定 企业在制定营销方案的时候所面临的最大问题就是把产品卖给“谁”？也就是确定目标客户群体的问题。 市场之大，消费者何其众也，国内尚且如此，更何况国际市场，企业在确定目标客户群体的时候，首先要针对所有的客户进行初步判别和确认。 在初步确定目标客户群体时，必须关注于企业的战略目标，它包括两个方面的内容，一方面是寻找企业品牌需要特别针对的具有共同需求和偏好的消费群体， 另一方面是寻找能帮助公司获得期望达到的销售收入和利益的群体。 通过分析居民可支配收入水平、年龄分布、地域分布、购买类似产品的支出统计，可以将所有的消费者进行初步细分，筛选掉因经济能力、地域限制、消费习惯等原因不可能
为企业创造销售收入的消费者，保留可能形成购买的消费群体，并对可能形成购买的消费群体进行某种一维分解，分解的标准 >标准可以依据年龄层次，也可以依据购买力水平，也可以依据有理可循的消费习惯。 由于分析方法更趋于定性分析，经过筛选保留下的消费群体的边界可能是模糊的，需要进一步的细化与探索。 目标客户群体的二次细分 在根据企业战略目标初步判别目标客户群体的轮廓之后，企业需要对这个范围较大的目标客户群体进行二次细分，目的是帮助企业确认目标客户群体的最终方案。 首先通过综合定性判别结合小规模的客户调查或经销商访谈，丰富已经初步确定的战略目标客户群体分解标准，赋值形成购买驱动/衰竭曲线，如以年龄层次、购买频率、购买支出占可支配收入的额度为分解标准赋值切等等。 衰竭型 购买驱动/衰竭曲线 购买驱动力 客户群体类型 客户类型B 客户类型C 客户类型A 旺盛型 驱动型 举例假设：如上图所示，被列如衰竭型的客户类型C应被排除在目标客户群体的最终方案外。 其次，需要对总体目标客户群体进行排序。即确定首要关注对象、次要目标和辐射人群。 首要关注对象是指在总体目标客户群体中，有最高消费潜力的那部分消费者；次要目标是指与企业战略目标有分歧的但能为产品创造重要销售机会的消费者；辐射人群是指处于总体目标客户群体内购买欲望最弱的那部分群体，但他们可以被企业的营销手段 影响而形成偶然购买甚至最终成为固定购买群体。 首要关注对象是企业在营销战略中最值得关注的群体，是在总体目标客户群体中具有最
统一身份认证平台讲解
统一身份认证平台设计方案 1）系统总体设计 为了加强对业务系统和办公室系统的安全控管，提高信息化安全管理水平，我们设计了基于PKI/CA技术为基础架构的统一身份认证服务平台。 1.1.设计思想 为实现构建针对人员帐户管理层面和应用层面的、全面完善的安全管控需要，我们将按照如下设计思想为设计并实施统一身份认证服务平台解决方案： 内部建设基于PKI/CA技术为基础架构的统一身份认证服务平台，通过集中证书管理、集中账户管理、集中授权管理、集中认证管理和集中审计管理等应用模块实现所提出的员工帐户统一、系统资源整合、应用数据共享和全面集中管控的核心目标。 提供现有统一门户系统，通过集成单点登录模块和调用统一身份认证平台服务，实现针对不同的用户登录，可以展示不同的内容。可以根据用户的关注点不同来为用户提供定制桌面的功能。 建立统一身份认证服务平台，通过使用唯一身份标识的数字证书即可登录所有应用系统，具有良好的扩展性和可集成性。 提供基于LDAP目录服务的统一账户管理平台，通过LDAP中主、从账户的映射关系，进行应用系统级的访问控制和用户生命周期维护
管理功能。 用户证书保存在USB KEY中，保证证书和私钥的安全，并满足移动办公的安全需求。 1.2.平台介绍 以PKI/CA技术为核心，结合国内外先进的产品架构设计，实现集中的用户管理、证书管理、认证管理、授权管理和审计等功能，为多业务系统提供用户身份、系统资源、权限策略、审计日志等统一、安全、有效的配置和服务。 如图所示，统一信任管理平台各组件之间是松耦合关系，相互支撑又相互独立，具体功能如下： a)集中用户管理系统：完成各系统的用户信息整合，实现用户生 命周期的集中统一管理，并建立与各应用系统的同步机制，简 化用户及其账号的管理复杂度，降低系统管理的安全风险。
统一身份认证权限管理系统
统一身份认证权限管理系统 使用说明
目录 第1章统一身份认证权限管理系统 (3) 1.1 软件开发现状分析 (3) 1.2 功能定位、建设目标 (3) 1.3 系统优点 (4) 1.4 系统架构大局观 (4) 1.5物理结构图 (5) 1.6逻辑结构图 (5) 1.7 系统运行环境配置 (6) 第2章登录后台管理系统 (10) 2.1 请用"登录"不要"登陆" (10) 2.2 系统登录 (10) 第3章用户（账户）管理 (11) 3.1 申请用户（账户） (12) 3.2 用户（账户）审核 (14) 3.3 用户（账户）管理 (16) 3.4 分布式管理 (18) 第4章组织机构（部门）管理 (25) 4.1 大型业务系统 (26) 4.2 中小型业务系统 (27) 4.3 微型的业务系统 (28) 4.4 内外部组织机构 (29) 第5章角色（用户组）管理 (30) 第6章职员（员工）管理 (34) 6.1 职员（员工）管理 (34) 6.2 职员（员工）的排序顺序 (34) 6.3 职员（员工）与用户（账户）的关系 (35) 6.4 职员（员工）导出数据 (36) 6.5 职员（员工）离职处理 (37) 第7章内部通讯录 (39) 7.1 我的联系方式 (39) 7.2 内部通讯录 (40) 第8章即时通讯 (41) 8.1 发送消息 (41) 8.2 即时通讯 (43) 第9章数据字典（选项）管理 (1) 9.1 数据字典（选项）管理 (1) 9.2 数据字典（选项）明细管理 (3) 第10章系统日志管理 (4) 10.1 用户（账户）访问情况 (5) 10.2 按用户（账户）查询 (5) 10.3 按模块（菜单）查询 (6) 10.4 按日期查询 (7) 第11章模块（菜单）管理 (1) 第12章操作权限项管理 (1) 第13章用户权限管理 (4) 第14章序号（流水号）管理 (5) 第15章系统异常情况记录 (7) 第16章修改密码 (1) 第17章重新登录 (1) 第18章退出系统 (3)
统一身份认证平台功能描述
统一身份认证平台功能 描述 文稿归稿存档编号：[KKUY-KKIO69-OTM243-OLUI129-G00I-FDQS58-
数字校园系列软件产品统一身份认证平台 功能白皮书 目录
1产品概述 1.1产品简介 随着校园应用建设的逐步深入，已经建成的和将要建成的各种数字校园应用系统之间的身份认证管理和权限管理出现越来越多的问题：用户需要记录多个系统的密码，经常会出现忘记密码的情况；在登 录系统时需要多次输入用户名/密码，操作繁琐。 各个系统之间的账号不统一，形成信息孤岛现象，导致学校管理工 作重复，增加学校管理工作成本。 新开发的系统不可避免的需要用户和权限管理，每一个新开发的系 统都需要针对用户和权限进行新开发，既增加了学校开发投入成 本，又增加了日常维护工作量 针对学生、教职工应用的各种系统，不能有效的统一管理用户信 息，导致学生在毕业时、教职工在离退休时不能及时地在系统中 清除这部分账号，为学校日后的工作带来隐患。 缺乏统一的审计管理，出现问题，难以及时发现问题原因。 缺乏统一的授权管理，出现权限控制不严，造成信息泄露。 统一身份认证平台经过多年的实践和积累，通过提供统一的认证服务、授权服务、集中管理用户信息、集中审计，有效地解决了以上问题，赢得客户的好评。
1.2应用范围 2产品功能结构 统一身份认证平台功能结构图 3产品功能 3.1认证服务 3.1.1用户集中管理 统一身份认证平台集中管理学校的所有教职员工和学生信息，所有的用户信息和组织机构信息存储在基于LDAP协议的OpenLDAP目录服务中，保证数据的保密性和读取效率。通过用户同步功能，及时地把关键业务系统中的用户信息同步到统一认证平台中，然后通过平台再分发给需要的业务系统，保证账号的一致性。 为所有的用户设置权限生效起止日期，即使不对用户做任何操作，在权限生效期外的用户也无法通过认证，保证了系统的安全性。 用户管理
统一身份认证平台功能描述
统一身份认证平台功 能描述 Revised on November 25, 2020
数字校园系列软件产品 统一身份认证平台 功能白皮书
目录
1产品概述 1.1产品简介 随着校园应用建设的逐步深入，已经建成的和将要建成的各种数字校园应用系统之间的身份认证管理和权限管理出现越来越多的问题： ?用户需要记录多个系统的密码，经常会出现忘记密码的情况；在登录系统时需要多次输入用户名/密码，操作繁琐。 ?各个系统之间的账号不统一，形成信息孤岛现象，导致学校管理工作重复，增加学校管理工作成本。 ?新开发的系统不可避免的需要用户和权限管理，每一个新开发的系统都需要针对用户和权限进行新开发，既增加了学校开发投入成本，又增加 了日常维护工作量 ?针对学生、教职工应用的各种系统，不能有效的统一管理用户信息，导致学生在毕业时、教职工在离退休时不能及时地在系统中清除这部分账 号，为学校日后的工作带来隐患。 ?缺乏统一的审计管理，出现问题，难以及时发现问题原因。 ?缺乏统一的授权管理，出现权限控制不严，造成信息泄露。 统一身份认证平台经过多年的实践和积累，通过提供统一的认证服务、授权服务、集中管理用户信息、集中审计，有效地解决了以上问题，赢得客户的好评。 1.2应用范围
2产品功能结构 统一身份认证平台功能结构图 3产品功能 3.1认证服务 3.1.1用户集中管理 统一身份认证平台集中管理学校的所有教职员工和学生信息，所有的用户信息和组织机构信息存储在基于LDAP协议的OpenLDAP目录服务中，保证数据的保密性和读取效率。通过用户同步功能，及时地把关键业务系统中的用户信息同步到统一认证平台中，然后通过平台再分发给需要的业务系统，保证账号的一致性。
统一身份认证平台讲解-共38页知识分享
统一身份认证平台讲解-共38页
统一身份认证平台设计方案 1）系统总体设计 为了加强对业务系统和办公室系统的安全控管，提高信息化安全管理水平，我们设计了基于PKI/CA技术为基础架构的统一身份认证服务平台。 1.1.设计思想 为实现构建针对人员帐户管理层面和应用层面的、全面完善的安全管控需要，我们将按照如下设计思想为设计并实施统一身份认证服务平台解决方案： 内部建设基于PKI/CA技术为基础架构的统一身份认证服务平台，通过集中证书管理、集中账户管理、集中授权管理、集中认证管理和集中审计管理等应用模块实现所提出的员工帐户统一、系统资源整合、应用数据共享和全面集中管控的核心目标。 提供现有统一门户系统，通过集成单点登录模块和调用统一身份认证平台服务，实现针对不同的用户登录，可以展示不同的内容。可以根据用户的关注点不同来为用户提供定制桌面的功能。 建立统一身份认证服务平台，通过使用唯一身份标识的数字证书即可登录所有应用系统，具有良好的扩展性和可集成性。
提供基于LDAP目录服务的统一账户管理平台，通过LDAP中主、从账户的映射关系，进行应用系统级的访问控制和用户生命周期维护管理功能。 用户证书保存在USB KEY中，保证证书和私钥的安全，并满足移动办公的安全需求。 1.2.平台介绍 以PKI/CA技术为核心，结合国内外先进的产品架构设计，实现集中的用户管理、证书管理、认证管理、授权管理和审计等功能，为多业务系统提供用户身份、系统资源、权限策略、审计日志等统一、安全、有效的配置和服务。 如图所示，统一信任管理平台各组件之间是松耦合关系，相互支撑又相互独立，具体功能如下：
18~25岁消费群体特征
18-25岁校园主流人群，这个群体的特征是高学历、高敏感度（指信息的整合能力和对新事物的接受能力强）、高消费潜力（指今后的消费能力潜力高）。他们不仅是众多产品和服务所瞄准的目标消费群体，还是诸多新产品、新服务的早期消费者，在消费的传播链条中具有相当的影响力。这个群体好比消费和媒体使用的风向标，他们所表现出来的消费或媒体使用特征往往预示着大众即将跟进的潮流。 消费特征：我们可以将这个年龄段的顾客群体称之为“不成熟型的消费群体”，因为18～2２岁之间的顾客通常还没有独立的经济来源。22～25岁之间的顾客刚刚参加工作，收入不稳定或较低，这使她们的消费水平受到一定的制约。但是，年轻的心态、较高的欣赏水平，以及毫无家庭责任困扰的消费习惯，都会促使这个年龄层的顾客成为最积极、最冲动的消费者。 这一年轻的消费群体，几乎不需要你费尽口舌和心力，只要产品是她们看中的，所需付的钞票不会多于当时她们钱包中的所有，她们都会毫不犹豫地购买。 18-20岁年轻人的价值为特征关键词∶“稳健”、“功利”、“求变”、“媒介”、“公益”、“社会”、“经济”、“公民”、“国际”、“中庸”、“自由”和”实惠”。结果也显示，18-20岁人群尽管具有比较“个人”化的意识，并且“求变”、“媒介”，希望自己的生活不断变化，喜欢流行、时髦的东西。同时，还有一部分相对保守性意识体现，如∶“经济”、“中庸”和“实惠”等。整体上反映了18-20年龄组的相当多元、多层价值意识特征。 21-23岁人群价值关键词∶“媒介”、“公民”、“经济”、“求变”、“官本”、“广告”、“中庸”、“名利”、“品味”、“社会”和“自我”。突出特点在于“媒介”和“公民”。这个年龄段相对于18-20岁人群多了一些“官本”、“广告”、“名利”和“品味”的表现，“经济”相对有所加强。面临生活、就业或者即将到来的各方面的压力，他们从因特网、广播、杂志等各种广告获取信息，“名利”、“官本”和“品味”意识反映出一定的社会生活“认知”。 24-26岁人群价值观关键词∶“稳健”、“求变”、“广告”、“自我”、“实惠”、“中庸”、“求新”、“媒介”、“国际”和“诚信”。“稳健”意识最为突出，其次为“求变”“广告”意识。这个年龄段也追求稳健，但是和18-20岁人群有所不同，他们的“稳”不是体现为“平平安安”，而是“我不喜欢负债消费”。 朵朵。可可S·朵系列：18-25岁，针对大中城市的年轻女性；高校在读女生或者事业刚起步的年轻白领，处于实现人生奋斗目标的起始阶段。是为懂得享受美好生活，渴望自由的时尚年轻女性提供的品牌家居服。
统一身份认证系统技术方案
智慧海事一期统一身份认证系统 技术方案
目录 目录...................................................................................................................................................... I 1.总体设计 (2) 1.1设计原则 (2) 1.2设计目标 (3) 1.3设计实现 (3) 1.4系统部署 (4) 2.方案产品介绍 (6) 2.1统一认证管理系统 (6) 2.1.1系统详细架构设计 (6) 2.1.2身份认证服务设计 (7) 2.1.3授权管理服务设计 (10) 2.1.4单点登录服务设计 (13) 2.1.5身份信息共享与同步设计 (15) 2.1.6后台管理设计 (19) 2.1.7安全审计设计 (21) 2.1.8业务系统接入设计 (23) 2.2数字证书认证系统 (23) 2.2.1产品介绍 (23) 2.2.2系统框架 (24) 2.2.3软件功能清单 (25) 2.2.4技术标准 (26) 3.数字证书运行服务方案 (28) 3.1运行服务体系 (28) 3.2证书服务方案 (29) 3.2.1证书服务方案概述 (29) 3.2.2服务交付方案 (30) 3.2.3服务支持方案 (36) 3.3CA基础设施运维方案 (38) 3.3.1运维方案概述 (38) 3.3.2CA系统运行管理 (38) 3.3.3CA系统访问管理 (39) 3.3.4业务可持续性管理 (39) 3.3.5CA审计 (39)
统一身份认证、统一系统授权、统一系统审计、统一消息平台、统一内容管理方案设计
基础支撑层 统一身份认证（SSO） 统一身份认证解决用户在不同的应用之间需要多次登录的问题。目前主要有两种方法，一种是建立在PKI，Kerbose和用户名/口令存储的基础上；一种是建立在cookie的基础上。统一身份认证平台主要包括三大部分：统一口令认证服务器、网络应用口令认证模块(包括Web 口令认证、主机口令认证模块、各应用系统口令认证模块等) 和用户信息数据库，具体方案如下图。 1、采用认证代理，加载到原有系统上，屏蔽或者绕过原有系统的认证。 2、认证代理对用户的认证在公共数据平台的认证服务器上进行，认证代理可以在认证服务器上取得用户的登录信息、权限信息等。 3、同时提供一个频道链接，用户登录后也可以直接访问系统，不需要二次认证。 4、对于认证代理无法提供的数据信息，可以通过访问Web Service接口来获得权限和数据信息。 单点登录认证的流程如下图所示：
单点登录只解决用户登录和用户能否有进入某个应用的权限问题，而在每个业务系统的权限则由各自的业务系统进行控制，也就是二次鉴权的思想，这种方式减少了系统的复杂性。统一身份认证系统架构如下图所示。 统一系统授权 统一系统授权支撑平台环境中，应用系统、子系统或模块统通过注册方式向统一系统授权支撑平台进行注册，将各应用系统的授权部分或全部地委托给支撑平台，从而实现统一权限管理，以及权限信息的共享，其注册原理如下图。
用户对各应用系统的访问权限存放在统一的权限信息库中。用户在访问应用系统的时候，应用系统通过统一授权系统的接口去查询、验证该用户是否有权使用该功能，根据统一系统授权支撑平台返回的结果进行相应的处理，其原理如下图。 统一系统授权支撑平台的授权模型如下图所示。在授权模型中采用了基于角色的授权方式，以满足权限管理的灵活性、可扩展性和可管理性的需求 块
目标消费人群(25岁-35岁)客户群分析
卡绚目标人群（25-36岁）客户群分析25-36岁 青年人类型收入类型 对产品的需 求类型 消费心理及行 为特征 购买动机 影响动机 因素 阅读网站的习 惯 女性
性
综上所述，针对青年白领这类消费群体我们公司产品在广告宣传时应该注意以下方面： （1）这类消费者追求前卫时尚，注重个性品味，在产品广告宣传时重点宣传产品的“轻生活”的时尚健理念和健康运动精神，体现产品时尚性的新概念主题，以迎合这类消费者对追求时尚健康生活方式的需求。（2）这类消费者中大多数人空闲时间少，工作压力大，在产品新的健身运动理念宣传的同时，也要着重突出产品的安全、便捷等特点，以满足广大青年白领对即时放松和休闲的需求。 （3）针对青年女性对塑身的强烈需求，要结合相关的运动学原理和相关权威数据宣传产品的科学性和便捷性。 （4）这类消费者大都受过良好的高等教育，掌握信息量大，不轻易受舆论及广告宣传的影响，在选择此类产品时比较理性，注重商品的功能性和科学性在进行产品广告宣传时要注意用词严谨，不可过分夸大产品的功效，以避免消费者对我们产品产生抵触心理 （5）这类消费者不但注重物质需求的满足，更注重精神心理需求的满足，在产品广告宣传时我们要不断挖掘消费者的情感需求，例如：在产品宣传时从消费者实际利益入手，尊重和理解消费者，让消费者产生归属感。 （6）这类消费者喜欢攀比，争强好胜，虚荣心强。在产品广告宣传时要注意利用消费者的攀比好胜、虚荣的心理，巧妙的刺激他们购买。 （7）在针对中下层白领进行产品宣传时要注重突出产品的物美价廉，实用性强，性价比高的特点。 （8）公司在产品广告宣传的同时也要不断提高服务水平和收集客户的反馈信息，重视客户提出来的问题，及时予以解决。 （9）这类消费者注重自我表现和个性化，在产品广告宣传时也要重点宣传产品的个性化定制服务，满足广大青年白领对自我表现和个性化的需求。 （10）这类消费者容易产生感性购买即冲动性购买，我们在产品广告宣传时要给消费者营造一个良好舒适的购物环境，例如：要时尚大气、精致美观，产品陈列要富有想象力，产品命名要新颖有创造性，产品图片要清晰美观、丰富全面。
统一身份认证系统
1.1. 统一身份认证系统 通过统一身份认证平台，实现对应用系统的使用者进行统一管理。实现统一登陆，避免每个人需要记住不同应用系统的登陆信息，包含数字证书、电子印章和电子签名系统。 通过综合管理系统集成，实现公文交换的在线电子签章、签名。 统一身份认证系统和SSL VPN、WEB SSL VPN进行身份认证集成。 2. 技术要求 ?基于J2EE实现，支持JAAS规范的认证方式扩展 ?认证过程支持HTTPS，以保障认证过程本身的安全性 ?支持跨域的应用单点登陆 ?支持J2EE和.NET平台的应用单点登陆 ?提供统一的登陆页面确保用户体验一致 ?性能要求：50并发认证不超过3秒 ?支持联合发文：支持在Office中加盖多个电子印章，同时保证先前加 盖的印章保持有效，从而满足多个单位联合发文的要求。 ?支持联合审批：支持在Office或者网页（表单）中对选定的可识别区 域内容进行电子签名，这样可以分别对不同人员的审批意见进行单独的电 子签名。 ? Office中批量盖章：支持两种批量签章方式： ?用户端批量盖章； ?服务器端批量盖章。 ?网页表单批量签章：WEB签章提供批量表单签章功能，不需要打开单个 表单签章，一次性直接完成指定批量表单签章操作，打开某一表单时，能 正常显示签章，并验证表单完整性。 ?提供相应二次开发数据接口：与应用系统集成使用，可以控制用户只能 在应用系统中签章，不能单独在WORD/EXCEL中签章，确保只有具有权限的人才可以签章,方便二次开发。 ?满足多种应用需求：电子签章客户端软件支持MS Office、WPS、永中 Office、Adobe PDF、AutoCAD等常用应用软件环境下签章，网页签章控件 或电子签章中间件则为几乎所有基于数据库的管理信息系统提供了电子签
统一身份认证与单点登录系统建设方案
福建省公安公众服务平台 统一身份认证及单点登录系统建设方案 福建公安公众服务平台建设是我省公安机关“三大战役”社会管理创新的重点项目之一；目前平台目前已经涵盖了公安厅公安门户网 站及网站群、涵盖了5+N服务大厅、政民互动等子系统；按照规划，平台还必须进一步拓展便民服务大厅增加服务项目，电子监察、微博监管等系统功能，实现集信息公开、网上办事、互动交流、监督评议 功能为一体的全省公安机关新型公众服务平台。平台涵盖的子系统众多，如每个子系统都用自己的身份认证模块，将给用户带来极大的不便；为了使平台更加方便易用，解决各子系统彼此孤立的问题，平台 必须增加统一身份认证、统一权限管理及单点登录功能。 一、建设目标 通过系统的建设解决平台用户在访问各子系统时账户、密码不统一的问题，为用户提供平台的统一入口及功能菜单；使平台更加简便易用，实现“一处登录、全网漫游”。同时，加强平台的用户资料、授权控制、安全审计方面的管理，确保用户实名注册使用，避免给群 众带来安全风险；实现平台各子系统之间资源共享、业务协同、互联 互通、上下联动；达到全省公安机关在线服务集成化、专业化的目标。 二、规划建议 统一身份认证及单点登录系统是福建公安公众服务平台的核心 基础系统；它将统一平台的以下服务功能：统一用户管理、统一身份 认证、统一授权、统一注册、统一登录、统一安全审计等功能。系统 将通过标准接口（WebService接口或客户端jar包或dll动态链接库）向各子系统提供上述各类服务；各业务子系统只要参照说明文档，做适当集成改造，即可与系统对接，实现统一身份认证及单点登录， 实现用户资源的共享，简化用户的操作。
掌握五种类型消费群体
掌握五种类型消费群体 泉露净水器教您掌握五种类型消费群体。在市场销售中，会经常遇到一些问题，比如错过了推荐一款高档机器的机会、或者是顾客被吓跑了、亦或者是无法与顾客进行有效沟通等等。泉露宏辅策划根据多年市场经验，对净水器消费者心理进行了归类，净水器目标消费群体大致可分为以下五种类型： 第一类是理智消费型。 这一类消费者普遍较年青，文化水平较高，而且积极上进，对个人的发展有很清晰的目标，有社会责任感。他们对于消费方面会比较注重品质和性价比，是相对理性而且务实的一类。他们对于电视、互联网和报纸的广告影响较大，喜欢社交，对省内游有很明显的偏好。 第二类属于经济消费型。 他们大都收入不高，学历低，分布在年青一代和老年一代的人群中。大都是普通职员和技术人员。他们接触最多的是电视、报纸和互联网广告。在消费方面，价格是他们消费的关键因素，并不太注重品牌。他们追求平静和轻松，不习惯被约束，寻求健康、有活力的生活方式。 第三类是中庸消费型。 对于消费，他们没有明显的消费心理偏好。大都是年青人，收入不高，文化水平也没有显著差异，受电视广告的影响会更大，对城市周边游和省内游的偏好接近。他们对任何事情都不表示强烈的好恶，生活和工作并没有大的波澜起伏。 第四类是时尚消费型。 这一类中青年人比较多，他们的收入水平高，文化水平高，中间阶层的比例也高。他们追求标新立异，喜欢时尚，喜欢引人注意，也注重健康，会定时的参加运动和锻炼，也喜欢休闲娱乐和社交活动。对于消费，他们旨在追求时尚和个人表现，容易受到广告和潮流的影响，而且对产品的质量和安全也较为关注。因此，他们对电视、杂志和报纸类的广告接触较多。如果出游，会对省外游的意向表现出向往。 最后一类是炫耀消费型。 他们不安于现状，在极限体验、冒险中寻求乐趣，也会努力尝试更多的生活方式。这一类人普遍收入水平高，年龄、文化和职业并没有显著的差异。他们接触电视、报纸和电台广播广告的机会较多。在消费方面，他们喜欢讲究个性和自我表现，而且具有符号消费特性，并不会太关注价格，喜欢购物和社交活动，外省游和港澳等海外游的机会相对较高。 净水器的目标消费群，大概就是以上五种类型。了解了目标消费者的这些特性之后，有针对性地做产品、做宣传、做服务，在终端销售中才能构建差异化的核心竞争力。
统一身份认证平台
统一身份认证平台 一、主要功能 1.统一身份识别； 2.要求开放性接口，提供源代码，扩展性强，便于后期与其他系统对接； 3.支持移动终端应用（兼容IOS系统、安卓系统；手机端、PAD端；） 4.教师基础信息库平台（按照教育信息化标准-JYT1001_教育管理基础代码实现） 5.学生基础信息库平台（按照教育信息化标准-JYT1001_教育管理基础代码实现） 二、系统说明 2.1单点登录：用户只需登录一次，即可通过单点登录系统（SSO）访问后台的多个应 用系统，无需重新登录后台的各个应用系统。后台应用系统的用户名和口令可以各不相同，并且实现单点登录时，后台应用系统无需任何修改。 2.2即插即用：通过简单的配置，无须用户修改任何现有B/S、即可使用。解决了当前 其他SSO解决方案实施困难的难题。 2.3多样的身份认证机制：同时支持基于PKI/CA数字证书和用户名/口令身份认证方式， 可单独使用也可组合使用。 2.4基于角色访问控制：根据用户的角色和URL实现访问控制功能。基于Web界面管 理：系统所有管理功能都通过Web方式实现。网络管理人员和系统管理员可以通过浏览器在任何地方进行远程访问管理。此外，可以使用HTTPS安全地进行管理。 三、系统设计要求 3.1业务功能架构 通过实施单点登录功能，使用户只需一次登录就可以根据相关的规则去访问不同的应用系统，提高信息系统的易用性、安全性、稳定性；在此基础上进一步实现用户在异构系统（不同平台上建立不同应用服务器的业务系统），高速协同办公和企业知识管理功能。 单点登录系统能够与统一权限管理系统实现无缝结合，签发合法用户的权限票据，从而能够使合法用户进入其权限范围内的各应用系统，并完成符合其权限的操作。 单点登录系统同时可以采用基于数字证书的加密和数字签名技术，对用户实行集中统一的管理和身份认证，并作为各应用系统的统一登录入口。单点登录系统在增加系统安全性、降低管理成本方面有突出作用，不仅规避密码安全风险，还简化用户认证的相关应用操作。 说明：CA安全基础设施可以采用自建方式，也可以选择第三方CA。 3.2具体包含以下主要功能模块： ①身份认证中心 ②存储用户目录：完成对用户身份、角色等信息的统一管理； ③授权和访问管理系统：用户的授权、角色分配；访问策略的定制和管理；用户授权信息 的自动同步；用户访问的实时监控、安全审计； ④身份认证服务：身份认证前置为应用系统提供安全认证服务接口，中转认证和访问请求； 身份认证服务完成对用户身份的认证和角色的转换； ⑤访问控制服务：应用系统插件从应用系统获取单点登录所需的用户信息；用户单点登 录过程中，生成访问业务系统的请求，对敏感信息加密签名； ⑥CA中心及数字证书网上受理系统：用户身份认证和单点登录过程中所需证书的签发； 四、技术要求 4.1技术原理 基于数字证书的单点登录技术，使各信息资源和本防护系统站成为一个有机的整体。 通过在各信息资源端安装访问控制代理中间件，和防护系统的认证服务器通信，利用系统提供的安全保障和信息服务，共享安全优势。 其原理如下： 1)每个信息资源配置一个访问代理，并为不同的代理分配不同的数字证书，用来保
(3)统一身份认证平台功能描述
数字校园系列软件产品 统一身份认证平台 功能白皮书
目录 1 产品概述............................................................................................................................... - 1 - 1.1 产品简介................................................................................................................... - 1 - 1.2 应用范围................................................................................................................... - 1 - 2 产品功能结构....................................................................................................................... - 2 - 3 产品功能............................................................................................................................... - 2 - 3.1 认证服务................................................................................................................... - 2 - 3.1.1 用户集中管理............................................................................................... - 2 - 3.1.2 认证服务....................................................................................................... - 3 - 3.2 授权服务................................................................................................................... - 3 - 3.2.1 基于角色的权限控制................................................................................... - 3 - 3.2.2 授权服务....................................................................................................... - 4 - 3.3 授权、认证接口....................................................................................................... - 4 - 3.4 审计服务................................................................................................................... - 4 - 3.5 信息发布服务........................................................................................................... - 5 - 3.6 集成服务................................................................................................................... - 5 - 3.6.1 应用系统管理............................................................................................... - 5 - 3.6.2 应用系统功能管理....................................................................................... - 6 - 3.6.3 应用系统操作管理....................................................................................... - 6 -
统一身份认证权限管理系统
` 统一身份认证权限管理系统 使用说明
目录 第1章统一身份认证权限管理系统 (3) 1.1 软件开发现状分析 (3) 1.2 功能定位、建设目标 (3) 1.3 系统优点 (4) 1.4 系统架构大局观 (4) 1.5物理结构图 (5) 1.6逻辑结构图 (5) 1.7 系统运行环境配置 (6) 第2章登录后台管理系统 (10) 2.1 请用"登录"不要"登陆" (10) 2.2 系统登录 (10) 第3章用户（账户）管理 (11) 3.1 申请用户（账户） (12) 3.2 用户（账户）审核 (14) 3.3 用户（账户）管理 (15) 3.4 分布式管理 (18) 第4章组织机构（部门）管理 (25) 4.1 大型业务系统 (26) 4.2 中小型业务系统 (26) 4.3 微型的业务系统 (27) 4.4 外部组织机构 (28) 第5章角色（用户组）管理 (29) 第6章职员（员工）管理 (32) 6.1 职员（员工）管理 (32) 6.2 职员（员工）的排序顺序 (32) 6.3 职员（员工）与用户（账户）的关系 (33) 6.4 职员（员工）导出数据 (34) 6.5 职员（员工）离职处理 (35) 第7章部通讯录 (37) 7.1 我的联系方式 (37) 7.2 部通讯录 (38) 第8章即时通讯 (39) 8.1 发送消息 (39) 8.2 即时通讯 (41) 第9章数据字典（选项）管理 (1) 9.1 数据字典（选项）管理 (1) 9.2 数据字典（选项）明细管理 (3) 第10章系统日志管理 (4) 10.1 用户（账户）访问情况 (4) 10.2 按用户（账户）查询 (5) 10.3 按模块（菜单）查询 (6) 10.4 按日期查询 (7) 第11章模块（菜单）管理 (1) 第12章操作权限项管理 (1) 第13章用户权限管理 (4) 第14章序号（流水号）管理 (5) 第15章系统异常情况记录 (7) 第16章修改密码 (1) 第17章重新登录 (1) 第18章退出系统 (3)
上海科技统一身份认证平台
上海科技统一身份认证平台操作手册（个人用户） 上海市科学技术委员会 二0一九年五月
目录 1引言 (2) 1.1编写目的 (2) 1.2用户对象 (2) 1.3登录方式 (2) 2个人用户注册说明 (3) 2.1个人注册 (3) 2.2登录系统 (4) 2.2.1账号登录 (4) 2.2.2短信登录 (5) 2.2.3忘记密码 (6) 3个人信息管理说明 (8) 3.1首页 (8) 3.2实名认证 (8) 3.3关联单位 (9) 3.4资料完善 (11) 4申请科技专家说明 (13) 4.1入库专家基本条件 (13) 4.2入库流程 (13) 4.3申请专家 (13) 4.4撤回申请 (17) 4.5查看专家详情 (18)
1引言 1.1编写目的 为了使个人用户能够准确、方便的使用本平台功能来注册个人信息及申请专家，特编写了《上海科技统一身份认证平台操作手册（个人）》。 1.2用户对象 科研人员、专家等。 1.3登录方式 登录网址https://www.wendangku.net/doc/051602861.html,，支持pc端填报。建议使用谷歌浏览器或微软Edge浏览器。
2个人用户注册说明 2.1个人注册 具体操作如下： 1.点击“个人/专家”，跳转至账号注册页面 2.手机号码：输入手机号码 3.短信验证码：点击“获取验证码”按钮，获取验证码 4.输入密码（新密码：8-20位、数字和字母组合、区分大小写） 5.确认密码，确保与密码的一致性 6.真实姓名：输入真实姓名 7.证件类型：选择证件类型 8.证件号码：输入证件号码 9.点击“注册”按钮，系统校验该手机号是否已注册 10.如已注册，系统提示账号已存在 11.如未注册，系统提示注册成功页面 12.点击“立即登录”按钮，打开登录页面
统一身份认证系统建设方案
统一身份认证系统建设方案 发布日期：2008-04-01 1.1 研发背景 随着信息技术的不断发展，企业已逐渐建立起多应用、多服务的IT 架构，在信息化建设中起到十分重要的作用。但是各信息系统面向不同管理方向，各有其对应的用户群体、技术架构、权限体系，限制了系统之间的信息共享和信息交换，形成的信息孤岛。同时，每一个信息系统的用户拥有不同的角色（职能），需要操作不同的系统，难以对其需要和拥有的信息和操作进行综合处理，限制信息系统效率的发挥。在这种背景下企业准备实施内网信息门户系统。其中统一身份管理系统是内网信息门户系统的一个重要组成部分。 统一身份管理将分散的用户和权限资源进行统一、集中的管理，统一身份管理的建设将帮助实现内网信息门户用户身份的统一认证和单点登录，改变原有各业务系统中的分散式身份认证及授权管理，实现对用户的集中认证和授权管理，简化用户访问内部各系统的过程，使得用户只需要通过一次身份认证过程就可以访问具有相应权限的所有资源。 1.2 组成架构 汇信科技与SUN公司建立了紧密合作关系，汇信科技推出的统一身份认证解决方案基于SUN公司的Sun Java System Identity Manager和Sun Java System Access Manager以及Sun Java System Directory Server实现。主要包括受控层、统一访问控制系统（统一认证服务器）、统一身份管理系统（统一身份管理服务器）、目录服务器。 受控层位于各应用服务器前端，负责策略的判定和执行，提供AGENT和API两种部署方式。 统一认证服务器安装统一身份认证系统（AM），主要提供身份认证服务和访问控制服务。 统一认证服务器安装统一身份管理系统（IM），主要实现身份配给、流程自动化、委任管理、密码同步和密码重置的自助服务。 目录服务器部署Sun Java System Directory Server,是整个系统的身份信息数据中心。 1.1 功能描述 1.1.1 实现“一次鉴权”（SSO） “一次鉴权（认证和授权）”是指建立统一的资源访问控制体系。用户采用不同的访问手段（如Intranet、PSTN、GPRS等）通过门户系统