An upper limb rehabilitation system fusing machine vision and virtual reality

Abstract

The application discloses a kind of upper limb rehabilitation system of fusion machine vision and virtual reality, including binocular camera, human skeleton recognition algorithm module, compensation detection module, action recognition module, exoskeleton upper limb rehabilitation robot, virtual reality platform.Exoskeleton upper limb rehabilitation robot is driven by mechanical arm to carry out rehabilitation training on affected side;Binocular camera is used to collect image;Human skeleton recognition algorithm module locates human area from input view image, obtains the coordinates of each joint;Compensation detection module processes and analyzes trunk data, and identifies compensation behavior in training process;Action recognition module is used to identify the action category of healthy side, as instruction control mechanical arm;Virtual reality platform includes multiple rehabilitation games, realizes interaction with game through mechanical arm, and generates evaluation report.The whole system realizes the accuracy, digitization, effectiveness, autonomy, scientization, interesting, data and standardization of rehabilitation training process.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of upper limb rehabilitation, in particular to an upper limb rehabilitation system integrating machine vision and virtual reality. BACKGROUND

[0002] Stroke is a disease with high mortality and high disability rate. With the increasing number of hemiplegic patients and the severe shortage of rehabilitation therapists, traditional manual physical therapy cannot meet the huge rehabilitation needs, and the wide application of robotic intelligent rehabilitation platform is imperative. However, in the process of upper limb rehabilitation robot assisted rehabilitation training, patients often produce compensatory movement, which not only reduces the rehabilitation efficiency, but also may learn abnormal behavior. The existing upper limb rehabilitation robot lacks monitoring and evaluation mechanism for compensatory movement of hemiplegic upper limb, cannot provide timely feedback and improve rehabilitation training, and cannot effectively inhibit compensatory movement, thereby affecting the effectiveness of the upper limb rehabilitation system. At the same time, the existing upper limb rehabilitation system is relatively single in form and simple in function, the training process is boring and tedious, the patient's self-participation is not high, and the enthusiasm is insufficient, which seriously affects the effect of rehabilitation treatment. SUMMARY

[0003] In order to at least solve one of the problems existing in the prior art, the present application discloses an upper limb rehabilitation system integrating machine vision and virtual reality.

[0004] To achieve the above-mentioned purpose, the present application provides an upper limb rehabilitation system integrating machine vision and virtual reality, which comprises a binocular camera, a human body skeleton recognition algorithm module, a compensation detection module, a motion recognition module, an exoskeleton upper limb rehabilitation robot and a virtual reality platform.

[0005] The binocular camera is used for real-time acquisition of input image data;

[0006] The human body skeleton recognition algorithm module is used for processing the image data obtained by the binocular camera, locating the human body region from the input image data, obtaining the three-dimensional coordinate information of each joint of the human body, and further obtaining the human body posture and motion information based on the joint data;

[0007] The compensation detection module acquires the spatial position information of the trunk joint of the human body by the human body skeleton recognition algorithm module, perceives the posture information of the trunk in the active motion training process of the patient through the classifier model, judges whether the compensatory behavior occurs, and reminds the patient through the virtual reality platform;

[0008] The motion recognition module identifies the motion type of the healthy side of the patient based on the deep learning model according to the motion information of the joint of the healthy side of the patient, as the self-training intention of the patient to control the exoskeleton upper limb rehabilitation robot to execute the corresponding motion, and realizes convenient and intelligent human-computer interaction;

[0009] The exoskeleton upper limb rehabilitation robot is used for wearing on the affected side of a patient to drive the upper limb to participate in rehabilitation training and rehabilitation games.

[0010] The virtual reality platform is used for storing patient information, and realizing interaction with the augmented virtual reality game by receiving position data of the exoskeleton rehabilitation robot, displaying a rehabilitation training scene, a real-time movement posture of the patient and a result of compensation evaluation, and interacting with the patient through visual display and voice prompt in the game process, and finally generating a game report.

[0011] Further, the binocular camera is installed above the display and is inclined downward by 30 degrees to face the training scene, effectively avoiding occlusion in the training process, the binocular camera collects environmental image data at 60 fps, and three-dimensional space information is obtained by using a binocular positioning principle, and the positioning accuracy can reach millimeter level, thereby ensuring the accuracy of the rehabilitation process.

[0012] Further, the human body skeleton recognition algorithm processes the image data collected by the binocular camera to obtain three-dimensional coordinate information of each joint in the human body skeleton model. In order to reduce the complexity of predicting the three-dimensional information of the joint and improve the real-time performance of the algorithm, the 2D human body skeleton recognition algorithm from bottom to top is used to predict the 2D pixel coordinates of the human body joints in the image, then the three-dimensional coordinates of each joint pixel point are measured by using the binocular positioning principle, and after filtering algorithm is used to remove abnormal values, the accurate space information of the human body joints can be obtained, and the complex picture is simplified to a digital model of joint coordinates, thereby realizing the digitization of the rehabilitation process.

[0013] Further, in the compensation detection module, the classifier model is a compensation recognition algorithm based on 1DCNN, including a spliced feature extractor and a classifier, the feature extractor includes three layers of 1DCNN, and the classifier includes five layers of full connection network.

[0014] Further, the input features of the classifier are position information of four joints on the human body trunk, the posture information of the patient in the training process is predicted by the classifier model, whether there is a compensation behavior is judged, and the patient is reminded to correct the abnormal posture in time through the virtual reality platform, and the compensation behavior includes any one or more of trunk rotation, trunk forward inclination and shoulder lifting.

[0015] Further, the action recognition module identifies the action type of the healthy side of the patient based on the motion information of the joints of the healthy side of the patient and a self-designed deep learning model, reflects the training intention of the patient, and the patient controls the mechanical arm according to the intention. The identifiable actions include common rehabilitation training actions and daily actions, thereby realizing the autonomy of the rehabilitation process.

[0016] Further, the exoskeleton upper limb rehabilitation robot is worn on the affected side of the patient, as the main body directly interacting with the patient, and can complete active training, passive training and active-passive training and other training modes.

[0017] Further, the virtual reality platform includes a plurality of active training games and passive training games, and the patient selects a rehabilitation game according to a rehabilitation prescription. The patient is in a virtual reality environment, and the upper computer completes game interaction and training tasks in the virtual reality environment by receiving position data of the affected limb of the exoskeleton upper limb rehabilitation robot. The virtual reality platform stores basic information and training information of the patient, and generates a game report after each training is completed. The virtual reality platform displays a rehabilitation training scene, a real-time movement posture of the patient and a result of compensation detection, interacts with the patient through visual display and voice prompts during the game process, has a reward and error correction mechanism, and assists the patient in accurately completing rehabilitation training under the condition of no therapist. The rehabilitation process is interesting, data-based and standardized.

[0018] Further, in the action recognition module, a three-parallel branch action recognition model is used to recognize the action.

[0019] The three-parallel branch action recognition model includes a three-parallel encoder and decoder structure, a fully connected network layer and a Soft_max module.

[0020] The internal structure of the encoder is that after the input information is processed by two layers of 1DCNN, multi-channel spatial and short-time sequence features are extracted, and then the features are input into a channel attention module to encode the attention degree information of each channel. The internal structure of the decoder is that the input information is first processed by a self-attention module to obtain the attention distribution of each part of the data, and then the data with attention information is decoded by two layers of LSTM to extract the time sequence information therein. The fully connected network layer is used to learn the mapping relationship between the features and the action categories, and to predict the score of each sample belonging to each category. The Soft_max module is used to calculate the possibility of the sample belonging to a certain category according to the category score.

[0021] Compared with the prior art, the present application can at least achieve the following beneficial effects:

[0022] (1) The application is applied to upper limb rehabilitation training process. The binocular camera collects input image data in real time; the human body skeleton recognition algorithm module obtains three-dimensional coordinate information of each joint of the human body, and based on the joint data, the human body posture and action information can be further obtained; the compensation detection module obtains the spatial position information of the trunk joint of the human body by using the human body skeleton recognition algorithm, and perceives the posture information of the trunk in the active movement training process of the patient, judges whether the compensation behavior occurs, and reminds the patient through the virtual reality platform; the action recognition module identifies the action information of the healthy side of the patient based on the movement information of the joint of the healthy side of the patient, and based on the deep learning algorithm, the action information of the healthy side of the patient is identified as the independent training intention of the patient, controls the corresponding action of the exoskeleton upper limb rehabilitation robot, and realizes convenient and intelligent human-computer interaction; the exoskeleton upper limb rehabilitation robot is worn on the affected side of the patient, drives the upper limb to carry out rehabilitation training, and participates in the rehabilitation game; the virtual reality platform stores the patient information, and realizes the interaction with the enhanced virtual reality game by receiving the position data of the exoskeleton rehabilitation robot, displays the rehabilitation training scene, the real-time movement posture of the patient and the compensation evaluation result, interacts with the patient through visual display and voice prompt in the game process, and finally generates a game report. The whole system realizes the accuracy, digitization, effectiveness, autonomy, scientization, interest, data and standardization of the rehabilitation training process through the convenient, intuitive and intelligent human-computer interaction mode.

[0023] (2) The application can timely judge whether the compensation behavior exists through the compensation detection module, so that the user can timely correct the action, and ensure the effectiveness of the upper limb rehabilitation system. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 It is a kind of whole structure schematic diagram of upper limb rehabilitation system fusing machine vision and virtual reality;

[0025] Figure 2 It is a kind of step block diagram of patient using virtual reality platform of upper limb rehabilitation system fusing machine vision and virtual reality;

[0026] Figure 3 It is a kind of three-way parallel branch action recognition model structure diagram of upper limb rehabilitation system fusing machine vision and virtual reality.

[0027] The figure includes: 1-binoocular camera, 2-human body skeleton recognition algorithm module, 3-compensation detection module, 4-action recognition module, 5-exoskeleton upper limb rehabilitation robot, 6-virtual reality platform;

[0028] 11-registered or logged in, 12-set mechanical arm, 13-rehabilitation game, 14-game record, 15-evaluation report;

[0029] 21-encoder, 22-decoder, 23-full connection network layer, 24-Soft_max module. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] like Figure 1 As shown, this invention provides an upper limb rehabilitation system that integrates machine vision and virtual reality, used in the upper limb rehabilitation training process. It includes a binocular camera 1, a human skeleton recognition algorithm module 2, a compensation detection module 3, a motion recognition module 4, an exoskeleton upper limb rehabilitation robot 5, and a virtual reality platform 6.

[0032] The binocular camera 1 is used to acquire input image data in real time; the human skeleton recognition algorithm module 2 is used to process the image data acquired by the binocular camera 1, locate the human body region from the input image data, obtain the three-dimensional coordinate information of each joint of the human body, and further obtain human posture and movement information based on the joint data, so as to realistically restore the movement state; the compensation detection module 3 obtains the spatial position information of the joints of the human torso based on the human skeleton recognition algorithm, and designs a classifier model based on machine learning algorithm. Through the classifier model, it perceives the posture information of the torso during the patient's active movement training, judges whether compensation behavior occurs, and reminds the patient through the virtual reality platform; the action recognition module 4... Based on human skeletal recognition algorithms, the system identifies information about the joints of the patient's healthy limbs. Using deep learning algorithms, it identifies the patient's healthy side movement information, using this as the patient's autonomous training intention to control the exoskeleton upper limb rehabilitation robot to perform corresponding movements, achieving convenient and intelligent human-computer interaction. The exoskeleton upper limb rehabilitation robot 5 is worn on the patient's affected side to drive the upper limb in rehabilitation training and participate in rehabilitation games. The virtual reality platform 6 stores patient information and, by receiving position data from the exoskeleton upper limb rehabilitation robot 5, interacts with the augmented virtual reality game, displaying rehabilitation training scenes, the patient's real-time movement posture, and compensation detection results. During the game, it interacts with the patient through visual displays and voice prompts, and finally generates a game report. The entire system achieves accuracy, digitization, effectiveness, autonomy, scientific rigor, fun, data-driven approach, and standardization of the rehabilitation training process through convenient, intuitive, and intelligent human-computer interaction.

[0033] In some embodiments of the present application, the binocular camera 1 is installed above the display, tilted downward by 30 degrees towards the training scene, which can effectively avoid occlusion during the training process. Preferably, the binocular camera 1 collects environmental image data at 60 fps and obtains three-dimensional spatial information using binocular positioning principles, with a positioning accuracy of millimeter level, ensuring the accuracy of the rehabilitation process.

[0034] In some embodiments of the present application, the human body skeleton recognition algorithm module 2 processes the image data collected by the binocular camera to obtain three-dimensional coordinate information of each joint in the human body skeleton model. In order to reduce the complexity of predicting the three-dimensional information of the joints and improve the real-time performance of the algorithm, a bottom-up 2D human body skeleton recognition algorithm is used to predict the 2D pixel coordinates of the human joints in the image, and then the three-dimensional coordinates of each joint pixel are measured using binocular positioning principles. After filtering algorithm to remove outliers, the accurate spatial information of the human joints can be obtained. In this way, the complex picture can be simplified into a digital model of joint coordinates, realizing the digitization of the rehabilitation process.

[0035] In some embodiments of the present application, the compensation detection module 3 predicts the posture information of the human body trunk using the three-dimensional coordinate information of the joints, and the input feature is limited to the position information of the four joints on the human body trunk, which are the top of the trunk, the bottom of the trunk, the right shoulder and the left shoulder. The 3D coordinates of the four trunk joints in the continuous 5 frames of data are taken as input, a classifier model is designed based on machine learning algorithm, and the posture information of the patient during the training process is predicted through the classifier model to determine whether there is compensation behavior such as trunk rotation, trunk forward leaning and shoulder lifting.

[0036] Preferably, the classifier model refers to a compensation recognition algorithm based on 1DCNN trained by the patient compensation motion data set collected by the hospital. The algorithm model is composed of a feature extractor composed of three layers of 1DCNN and a classifier composed of five layers of fully connected network, and the trained classifier model can realize the classification and recognition of trunk compensation motion, and can remind the patient to correct the abnormal posture in time through the virtual reality platform, realizing the effectiveness of the rehabilitation process.

[0037] The action recognition module 4 identifies the action type of the healthy side of the patient based on the motion information of the joints of the healthy side of the patient based on a deep learning model, reflects the training intention of the patient, and the patient controls the mechanical arm according to his own intention to realize the self-control of the rehabilitation process. In some embodiments of the present application, the recognizable action types include common rehabilitation training actions and daily actions, such as standing still, abduction, forward stretching, pointing to the nose, wiping the face, picking up objects, turning over books, and combing the hair.

[0038] In some embodiments of the present application, the exoskeleton upper limb rehabilitation robot 5 is worn on the affected side of the patient, as the main body directly interacting with the patient, and can complete active training, passive training and active-passive training and other training modes, and realize the scientization of the rehabilitation process.

[0039] The virtual reality platform 6 stores a plurality of active training games and passive training games, and the patient selects a rehabilitation game according to a rehabilitation prescription. The patient is in a virtual reality environment, and the upper computer completes game interaction and training tasks in the virtual reality environment by receiving the position data of the affected limb of the exoskeleton upper limb rehabilitation robot 5. The virtual reality platform 6 stores the basic information and training information of the patient, and generates a game report after each training is completed. The virtual reality platform displays the rehabilitation training scene, the real-time movement posture of the patient and the results of compensation detection, interacts with the patient through visual display and voice prompts during the game process, has a reward and error correction mechanism, and assists the patient to accurately complete the rehabilitation training under the condition of no therapist. The rehabilitation process is interesting, data-based and standardized.

[0040] In addition, the present application also provides a virtual reality platform convenient to operate, referring to Figure 2 , the steps of using the virtual reality platform include: logging in or registering 11, setting the mechanical arm 12, rehabilitation game 13, game record 14 and evaluation report 15.

[0041] The login or registration 11 step registers the patient using the system for the first time, and after successful registration, the patient can log in, and subsequent use only needs to log in. The login can be through the user account and password, or through the mobile phone to scan the two-dimensional code for convenient authentication login.

[0042] The setting mechanical arm 12 step specifically includes power-on initialization, setting IP and port number, and selecting training. Before starting training, the mechanical arm needs to be powered on and initialized, and the motor is started; then the IP and port number are set, the connection mode is determined, the “connection” button is clicked, the mechanical arm program is connected through TCP communication; after the connection is completed, the training mode is selected according to the patient's condition, first select the left hand and the right hand, then select the active, passive, power assistance, impedance and other modes, input the security password after confirming the mode, confirm the selection, and after exiting each mode, the exoskeleton upper limb rehabilitation robot will restore the initialization position.

[0043] The rehabilitation game 13 step includes patient selection game type, game, is the core of virtual reality center module. The patient selects the game type according to the classification of the game center, selects the initiative, passive according to the action type, can select the shoulder joint, elbow joint, wrist joint, multi-joint, all according to the joint type, can choose simple, medium, difficult according to the difficulty, click the corresponding classification, there are multiple games to choose from, such as hitting the ground mouse, gold miner, home scene follow-up practice, etc.;The patient selects the corresponding classification according to the rehabilitation game diagnosis and treatment sheet, starts the game, and can immerse in the interactive game with the help of the mechanical arm.

[0044] The game record 14 step is that the patient can view his own game history after each game, including the action type, difficulty, training date, score and other specific contents, if the visual score is selected, the training results of the patient will be intuitively displayed in the form of chart.

[0045] The evaluation report 15 step is the evaluation report generated according to the doctor's diagnosis and the game training result, including various scale evaluation of the patient's motor ability and cognitive ability and other basic abilities, which can be selected as a visual table, which will draw the patient's evaluation results into a chart form.

[0046] In addition, in the action recognition module 4, a three-way parallel branch action recognition model is used to recognize the action. Referring to Figure 3 , the three-way parallel branch action recognition model includes an encoder 21, a decoder 22, a fully connected network layer 23 and a Soft_max module 24.

[0047] Further, the internal structure of the encoder 21 includes a 1D CNN layer and a channel attention module, wherein the convolutional layer is compressed from the input 1 channel length 240 original data to 10 channel length 4 feature data by two layers of 1D CNN;The channel attention module includes two convolutional layers after maximum pooling operation, which realizes the Squeeze and Excitation operation in SENet to obtain the channel attention information. After the input information is processed by the two layers of 1D CNN layer, the multi-channel spatial and short time sequence features are extracted, and then the features are input into the channel attention module to encode the attention information of each channel.

[0048] Further, the internal structure of the decoder 22 includes a self-attention module and an LSTM layer, wherein the self-attention module includes three linear network layers, which respectively calculate the Q matrix, K matrix and V matrix of the input data, and then obtain the attention information through two-step matrix multiplication operation;The LSTM layer includes two layers of LSTM network stacked, and finally outputs 10 channel length 8 feature data. In the decoding process, the input information is first processed by the self-attention module to obtain the importance of each part of the data, and then the data with attention information is decoded by two layers of LSTM layer to extract the time sequence information.

[0049] Further, the full connection network layer 23 contains multiple layers of linear neurons as a classifier, which accepts the extracted spatio-temporal features and automatically learns the mapping relationship between the features and the action categories through the powerful fitting ability of the deep network, to predict the score of each sample belonging to each category.

[0050] Further, the Soft_max module 24 can pull the large distance of the gap to be larger, which is used to calculate the possibility of the sample belonging to a certain category according to the category score of the classifier, to realize the prediction of a certain action.

[0051] The action recognition model adopts a three-way parallel structure to process the input action data, and splices the features extracted by the three branches to serve as the input data of the classifier, which can effectively improve the accuracy of action recognition compared with the existing model.

[0052] The action recognition model is designed for the significant temporal and spatial features in human actions, and is based on a hybrid encoding-decoding model of convolutional neural network CNN and long short-term memory unit LSTM, and adds an attention mechanism and adopts a three-way parallel branch structure as optimization, which can effectively extract the spatio-temporal features in the original data and improve the classification performance of the model on limb actions.

[0053] The above disclosure and teaching of the specification can also be changed and modified by those skilled in the art to which the present application belongs. Therefore, the present application is not limited to the specific embodiments disclosed and described above, and some modifications and changes of the present application should also fall within the protection scope of the claims of the present application.

Claims

1. A rehabilitation system for upper limbs that fuses machine vision and virtual reality, characterized in that, The system comprises a binocular camera (1), a human skeleton recognition algorithm module (2), a compensation detection module (3), a motion recognition module (4), an exoskeleton upper limb rehabilitation robot (5), and a virtual reality platform (6). The binocular camera (1) is used for real-time collection of input image data. The human skeleton recognition algorithm module (2) is used for processing of image data obtained by the binocular camera (1), locating a human body region from the input image data, obtaining three-dimensional coordinate information of each joint of the human body, and further obtaining human body posture and motion information based on the joint data. The compensation detection module (3) obtains spatial position information of a human body trunk joint by the human skeleton recognition algorithm module (2), perceives posture information of the trunk in a patient's active motion training process by a classifier model, judges whether compensation behavior occurs, and reminds the patient through the virtual reality platform (6). The motion recognition module (4) recognizes a motion type of a healthy side of the patient based on a deep learning model according to motion information of a joint of a healthy limb of the patient, takes the motion type as a self-training intention of the patient to control the exoskeleton upper limb rehabilitation robot (5) to perform a corresponding motion, and realizes convenient and intelligent human-computer interaction. The exoskeleton upper limb rehabilitation robot (5) is used for wearing on a diseased side of the patient to drive the upper limb to perform rehabilitation training and participate in a rehabilitation game. The virtual reality platform (6) is used for storage of patient information, realization of interaction with an augmented virtual reality game by receiving position data of the exoskeleton rehabilitation robot (5), display of a rehabilitation training scene, a real-time motion posture of the patient, and a result of compensation evaluation, interaction with the patient through visual display and voice prompt in a game process, and generation of a game report. In the motion recognition module, a three-parallel-branch motion recognition model is used to recognize the motion, three-parallel-branch structures are used to process input motion data, and features extracted by the three branches are spliced as input data of a classifier. The three-parallel-branch motion recognition model comprises a full connection network layer, a Soft_max module, and a three-parallel-branch encoder and decoder structure. The internal structure of the encoder is that input information is processed by two layers of 1DCNN to extract multi-channel spatial and short-time sequence features, and then the features are input into a channel attention module to encode attention information of each channel. The internal structure of the decoder is that input information is processed by a self-attention module to obtain attention distribution of each part of the data, and then the data with the attention information is decoded by two layers of LSTM to extract time sequence information. The full connection network layer is used to learn a mapping relationship between features and motion categories, and to predict scores of each sample belonging to each category. The Soft_max module is used to calculate a possibility of a sample belonging to a category according to category scores. 2.The upper limb rehabilitation system integrating machine vision and virtual reality according to claim 1, wherein: The binocular camera (1) is installed above a display and is inclined downward by 30 degrees towards a training scene. 3.The upper limb rehabilitation system integrating machine vision and virtual reality according to claim 1, wherein: In the human skeleton recognition algorithm module (2), a 2D human skeleton recognition algorithm from bottom to top is used to predict 2D pixel coordinates of human joints in an image. Then, three-dimensional coordinates of each joint pixel point are measured by using a binocular positioning principle. After filtering algorithm is used to remove abnormal values, accurate spatial information of the human joints can be obtained.

4. The upper limb rehabilitation system fusing machine vision and virtual reality according to claim 1, characterized in that: In the compensation detection module (3), the classifier model is a 1DCNN-based compensation identification algorithm, including a spliced feature extractor and a classifier, the feature extractor includes three layers of 1DCNN, and the classifier includes five layers of full connection network.

5. The upper limb rehabilitation system fusing machine vision and virtual reality according to claim 1, characterized in that: The input features of the classifier are the position information of four joints on the human body trunk, the posture information of the patient during the training process is predicted through the classifier model, it is judged whether there is a compensation behavior, and the patient is reminded to correct the abnormal posture in time through the virtual reality platform (6), and the compensation behavior includes whether there is any one or more of trunk rotation, trunk forward inclination and shoulder lifting. 6.The upper limb rehabilitation system integrating machine vision and virtual reality according to claim 1, wherein: The action type includes common rehabilitation training actions and daily actions.

7. The upper limb rehabilitation system fusing machine vision and virtual reality according to claim 1, characterized in that: The rehabilitation training includes active training, passive training and active-passive training modes. 8.The upper limb rehabilitation system integrating machine vision and virtual reality of claim 1, wherein: In the virtual reality platform (6), a plurality of active training games and passive training games are stored, the patient selects a rehabilitation game according to a rehabilitation prescription, the patient is in a virtual reality environment, and the upper computer completes game interaction and training tasks in the virtual reality environment by receiving position data of the affected limb of the exoskeleton upper limb rehabilitation robot (5).

Images