Medical care gait recognition and prediction method and system based on sequence model and medium
Through the gait recognition prediction method based on sequence model, the gait video image is processed and predicted by using convolutional neural network and Mamba model, the problem of insufficient gait prediction in the existing technology is solved, and the effective gait sequence is achieved Identification and condition risk assessment provide real-time early warning information.
Patent Information
- Application Number
- CN202410961821.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-12-13
AI Technical Summary
The existing gait recognition prediction methods mainly focus on wearable devices and visual image recognition. There is a lack of prediction methods based on sequence models, making it difficult to effectively predict gait sequences and related risk assessments in a certain time period.
Using a medical gait recognition prediction method based on sequence model, the gait video images are standardized, the foreground segmentation and recognition prediction are used to train the gait sequence types to identify and predict gait sequence types and match them with the disease database. Obtain the condition category and level.
The processing of low-resolution video image sequences is realized, and it can be applied to monitor the walking status of a person in a larger physical space, providing real-time early warning information, which is especially suitable for the prediction of the elderly in medical and nursing communities.
Smart Images
Figure CN119131881A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of computer vision technology and biometric identification technology, and in particular to a medical gait recognition prediction method, system and medium based on a sequence model. Background Art
[0002] Gait refers to the way a person walks. Conventional gait recognition refers to the identification of people based on their walking posture. In specific application scenarios, gait recognition can also refer to the specific identification of a certain walking posture, which is equivalent to the classification of posture. Gait classification is the classification of walking postures. Abnormal gait refers to abnormal movements and postures when walking. Abnormal gait can be used as an important biometric feature to identify and predict a person's biological condition, such as symptom detection. Gait prediction is the prediction of the posture in the next time period based on the current walking posture, as well as the judgment and detection of certain physical conditions reflected by this series of postures.
[0003] There are two main categories of existing gait recognition and prediction methods. One is based on wearable devices, using portable sensors such as pressure sensors and accelerometers to obtain digital features of key parts of the movement, and then obtain gait recognition and prediction by analyzing the features. Typical scenarios are rehabilitation hospitals and rehabilitation centers. The other is based on visual images and corresponding recognition algorithms. Typical scenarios are community monitoring, especially nursing home monitoring. From a research perspective, there are currently more studies on gait recognition, and relatively fewer studies on prediction based on gait recognition. In summary, gait prediction based on visual features has a larger application space and is also the focus of gait prediction research. Summary of the invention
[0004] The technical problem solved by the present invention is to provide a medical gait recognition and prediction method based on a sequence model, which recognizes the gait of the person in the currently captured video image and predicts the gait sequence in the next period of time through a comparative analysis algorithm. The risks associated with the gait are evaluated and suggestions and feedback are given in a timely manner.
[0005] The technical solution adopted by the present invention to solve the technical problem is: a medical gait recognition and prediction method based on a sequence model, comprising the following steps:
[0006] Gait video image acquisition steps: collecting gait motion video data from different perspectives to obtain a gait image sequence;
[0007] Gait image data preprocessing steps: standardize the images in the gait image sequence;
[0008] Foreground segmentation step of gait image task: obtain the character outline image sequence through the foreground segmentation model;
[0009] Gait recognition prediction step: Based on the gait recognition prediction model of the convolutional neural network and the sequence model, the recognition model and the prediction model are trained respectively, the gait sequence type data is obtained through the recognition model, and the type of the gait sequence diagram is output through the prediction model;
[0010] Result output step: Match the type of gait sequence diagram with the disease database to obtain the disease category and level.
[0011] Furthermore, in the step of acquiring the gait video image, specifically, the video image is acquired, and the video image is framed at a predetermined threshold interval to acquire a gait image sequence G = {g1, g2, ...g n}, where G represents the gait image sequence, g i Represents one frame of a gait image in a gait image sequence.
[0012] Furthermore, in the gait image data preprocessing step, the images in the gait image sequence are standardized, specifically:
[0013] Standardize the input video images and remove images where the proportion of people is too large or too small;
[0014] Subtract the redundant environment space that does not contain the person from the remaining image after culling;
[0015] Resizes an image to a defined uniform pixel resolution and uniform size.
[0016] Furthermore, the foreground segmentation step of the gait image task is specifically as follows:
[0017] The standardized images are divided into a training set, a test set and a validation set and input into the image segmentation model for image segmentation model training;
[0018] The trained model detects the video image, outputs labels and masks in sequence, traverses the detection results, converts the instance segmentation mask into a binary contour image, and refines the binary contour image.
[0019] Furthermore, in the gait recognition prediction step, the recognition model is trained, specifically, the binary contour image sequence is input into the YOYO detection model training, and a plurality of input gait category sequences I are output at the same time, where I = {x1, x2, ..., x n1}, x1 is the gait category.
[0020] Furthermore, in the gait recognition prediction step, the prediction model is trained, specifically, the gait category sequence I is input into the Mamba model for prediction model training, and the predicted gait category sequence T is output, T = {z1, z2, ..., z n2}, and n1>>n2, and n2≥1.
[0021] Furthermore, in the result output step, the type of the gait sequence diagram is matched with the disease database to obtain the disease category and level, specifically:
[0022] Establishing a disease data set, wherein the disease data set includes gait data and levels corresponding to different diseases;
[0023] Match the predicted gait category sequence T with the disease dataset;
[0024] Output the disease with the highest matching probability and its level.
[0025] The present invention also discloses a medical gait recognition and prediction system based on a sequence model, comprising a camera, a gait video image acquisition module, a gait image data preprocessing module, a gait image task foreground segmentation module, a gait recognition and prediction module and a result output module;
[0026] The camera is used to capture gait video images;
[0027] The gait video image acquisition module is used to acquire gait motion video data from different viewing angles captured by a camera to obtain a gait image sequence;
[0028] The gait image data preprocessing module is used to perform standardization processing on the images in the gait image sequence;
[0029] The foreground segmentation module of the gait image task is used to obtain a character outline image sequence through a foreground segmentation model;
[0030] The gait recognition prediction module is used for the gait recognition prediction model based on the convolutional neural network and the sequence model. The recognition model and the prediction model are trained respectively, the gait sequence type data is obtained through the recognition model, and the type of the gait sequence diagram is output through the prediction model;
[0031] The result output module is used to match the type of gait sequence diagram with the disease database to obtain the disease category and grade.
[0032] The present invention also discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the medical gait recognition and prediction method based on the sequence model are implemented.
[0033] The present invention also discloses a computer device, characterized in that it comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; wherein:
[0034] The memory is used to store computer programs;
[0035] The processor is used to execute the steps of the medical gait recognition and prediction method based on the sequence model by running the program stored in the memory.
[0036] The beneficial effects of the present invention are as follows: the method uses a fixed camera in a building or a following service robot to obtain a video of a person walking in a physical space and converts it into a sequence of images. The segmentation model is used to identify and classify the abnormal gait of the research object, and the Mamba sequence model is introduced to predict the gait, and based on this, the associated possible diseases are predicted. Therefore, the method can be used to process low-resolution video image sequences and can be applied to monitoring the walking status of people in a larger physical space. Furthermore, it can be used in a nursing home for medical care, and can be used to assist in the prediction of the elderly's diseases, so as to provide real-time early warning information to medical staff. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 Schematic diagram of the process of the medical gait recognition and prediction method based on the sequence model according to an embodiment of the present application.
[0038] Figure 2 Schematic diagram of the input and output of video image data passing through the three models of the system in sequence. DETAILED DESCRIPTION
[0039] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.
[0040] like Figure 1 As shown, the embodiment of the present application discloses a medical gait recognition and prediction method based on a sequence model, which is characterized by comprising the following steps:
[0041] Gait video image acquisition steps: collecting gait motion video data from different perspectives to obtain a gait image sequence;
[0042] Gait image data preprocessing steps: standardize the images in the gait image sequence;
[0043] Foreground segmentation step of gait image task: obtain the character outline image sequence through the foreground segmentation model;
[0044] Gait recognition prediction step: Based on the gait recognition prediction model of the convolutional neural network and the sequence model, the recognition model and the prediction model are trained respectively, the gait sequence type data is obtained through the recognition model, and the type of the gait sequence diagram is output through the prediction model;
[0045] This method first uses a camera fixed to a building or a camera attached to a follow-up service robot to capture gait video images of target persons from different perspectives in real time within the community. Before training the model, the data set is first preprocessed. Then, an image contour generation model, a gait recognition model, and a gait prediction model are constructed in sequence, and parameters are set for the model for training. The video shot on site is input into the above training model to obtain one or more subsequent predicted gait types. Finally, the gait type is matched with a disease database to obtain the disease category and level, and fed back to medical staff for further medical care measures.
[0046] This method uses a segmentation model, a recognition model, and a prediction model to identify, classify, and predict the abnormal gait of the research subject, and based on this, predicts the possible associated diseases, thereby facilitating medical staff to make disease judgments.
[0047] Specifically, in the step of acquiring the gait video image, the video image is acquired and the video image is framed at a predetermined threshold interval to acquire a gait image sequence G = {g1, g2, ...g n}, where G represents the gait image sequence, g i Represents one frame of a gait image in a gait image sequence.
[0048] Specifically, in this method, a camera fixed to a building or a camera attached to a following service robot is used to obtain gait motion video data from different viewing angles, which is real gait video image data.
[0049] The GEI_CASIA_B gait dataset can be used in the gait acquisition operation. GEI_CASIA_B is a widely used earlier cross-view gait dataset with multiple samples and multiple perspectives. In reality, in the process of acquiring the dataset, the characters in the scene are shot cross-view. The form refers to CASIA_B. In various scenes, it fully includes 11 perspectives (0°, 18°, ..., 180°). If the data associated with a certain disease in the later stage of the video image sequence is not easy to obtain in reality, and it faces problems such as individual privacy. Part of the later sequence data can also be simulated and drawn manually.
[0050] Specifically, a video is acquired through a camera, and the video is framed at a predetermined threshold interval to acquire an image sequence. For example, a video image sequence is acquired at a sampling frequency of 30 frames per second. The specific interval threshold can be set according to actual conditions.
[0051] In this embodiment, in the gait image data preprocessing step, the images in the gait image sequence are standardized, specifically:
[0052] Standardize the input video images and remove images where the proportion of people is too large or too small;
[0053] Subtract the redundant environment space that does not contain the person from the remaining image after culling;
[0054] Resize the image to a uniform pixel resolution and size, such as 64×44.
[0055] Specifically, the above-mentioned elimination of some images in which the proportion of the person in the image is too large or too small, the too large image refers to the image that cannot show the full body image, the too small image refers to the image with unclear outline, or for example, the image that accounts for less than 10% of the area, the specific requirements can be set by yourself.
[0056] The standardized data set is divided into training set, validation set and test set according to a certain ratio, such as 80%, 10%, 10%, etc., for subsequent model training.
[0057] In this embodiment, the foreground segmentation step of the gait image task is specifically as follows:
[0058] The standardized images are divided into a training set, a test set and a validation set and input into the image segmentation model for image segmentation model training;
[0059] The trained model detects the video image, outputs labels and masks in sequence, traverses the detection results, converts the instance segmentation mask into a binary contour image, and refines the binary contour image.
[0060] The aforementioned on-site real-shot data set and the data set produced by artificial simulation are divided into a training set, a test set and a validation set, and are randomly divided into a training set and a test set according to a certain ratio.
[0061] Then, you can use models with better segmentation effects such as MaskR-CNN and UNet++ to perform instance segmentation of people in the image.
[0062] Among them, MaskR-CNN is often used for pedestrian detection and segmentation, and gait recognition and prediction also require pedestrian detection and segmentation first, so MaskR-CNN is selected as the segmentation model here.
[0063] Using the simple, effective and stable Mask R-CNN instance segmentation framework, we can perform instance segmentation on video images to generate high-quality segmentation masks for the image, whether it is one person or multiple people, and obtain the foreground map of the pixel-level edges.
[0064] It should be explained that part of the training dataset comes from a public dataset, which already has the format of contour image sequence and contains mask annotations.
[0065] The field collection dataset and artificial simulation dataset are manually masked, and multiple pedestrian instances are labeled with one color. MaskR-CNN is based on FasterR-CNN. FasterR-CNN is used to predict potential target boxes and classification scores in images, while MaskR-CNN adds a branch to predict instance segmentation masks.
[0066] In actual operation, the existing MaskR-CNN model of the PyTorch framework and its training and testing parameters are used for model training. The trained model detects the input video image and outputs labels, segmentation image masks, etc. The detection results are traversed, and the segmentation image masks of the instance segmentation are converted into binary contour images, and further contour refinement is performed.
[0067] In this embodiment, in the gait recognition prediction step, the recognition model is trained, specifically: the binary contour image sequence is input into the YOYO detection model training, and a plurality of input gait category sequences I are output at the same time, where I = {x1, x2, ..., x n1}, x1 is the gait category.
[0068] Specifically, the steps of recognition model training include: binary image input, gait feature extraction, model parameter setting, and model training. The gait recognition prediction output is the gait category label.
[0069] Specifically, when training the recognition model, public data sets will be used for pre-training to obtain a basic training model, and then the actual collected data will be input for continuous training. The training data set of each model is divided into three parts: training set, validation set and test set. In this step, the test data set is used to evaluate the optimal parameters of the trained convolutional neural network.
[0070] The recognition and classification for gait prediction is first a multi-objective classification of the gait itself. The recognized multi-objective gait has a time series and is used as the input of the subsequent prediction model.
[0071] The recognition model used in this step is the existing YOLO series model. The YOLO series is also a target detection algorithm based on CNN. The gait classification in this step belongs to the classification of multiple targets and has multiple variations. Therefore, for the recognition and classification of such targets, a detection model based on the latest YOLOv9 architecture is built. According to the hardware and computing power, and considering the deployment on different servers, the appropriate version of YOLOv9 is selected according to the size of the data set.
[0072] Specifically, the above public data set is used to perform pre-training in the YOLOv9 model, and then continuous training is performed based on the actual collected data to obtain a continuously optimized recognition model.
[0073] In this embodiment, in the gait recognition prediction step, the prediction model is trained, specifically: the gait category sequence I is input into the Mamba model for prediction model training, and the predicted gait category sequence T is output, T = {z1, z2, ..., z n2}, and n1>>n2, and n2≥1.
[0074] Specifically, in this step, Figure 2 N+ in the figure is the next or several predicted gait types. The model used in this step is the Mamba model. Since gait data is a typical discrete sequence data composed of continuously changing moving targets, the Mamba model is a new sequence modeling model whose serialization representation structure is similar to that of RNN. Unlike RNN, Mamba directly uses linear transformation when calculating the output, instead of using activation function for nonlinearization. Compared with RNN, Mamba can be calculated in parallel, and has higher efficiency in training and reasoning than the Transformer model. The most important part of the Mamba model architecture is the state space model (SSM). Based on the structured state space model (SSM), the training and reasoning efficiency is improved by introducing the selective state space model (SSM). SM is good at capturing long-range dependencies, which can sequentially locate multiple output items of the model in long sequence data, that is, infer a certain type.
[0075] When training the model, the structure and parameters of the selective state space model (SSM), including the size of the state vector and the control matrix, are defined in turn. The training data set is used for training, and the model parameters are updated by the back propagation algorithm. According to the input gait category sequence I, the predicted gait category sequence T is obtained, where T = {z1, z2, ..., z n2}, and n1>>n2, and n2≥1.
[0076] The above-mentioned Mamba model is an existing model, and the specific details of how to train and use the model will not be repeated here.
[0077] In this method, a CNN-based multi-classification recognition model and a Mamba-based time series prediction model are used to accurately and quickly obtain the subsequent state of the gait, so as to obtain the probability of the disease by matching the predicted gait type with the disease database.
[0078] In this embodiment, in the final result output, the type of the gait sequence diagram is matched with the disease database to obtain the disease category and level, specifically:
[0079] Establishing a disease data set, wherein the disease data set includes gait data and levels corresponding to different diseases;
[0080] Match the predicted gait category sequence T with the disease dataset;
[0081] Output the disease with the highest matching probability and its level.
[0082] The final matched symptoms and their levels will be fed back to medical staff.
[0083] This method collects videos of people walking in space and converts them into sequence images. The segmentation model is used to identify and classify the abnormal gait of the research object, and the Mamba sequence model is introduced to predict the gait, and based on this, the possible associated diseases are predicted. Therefore, this method can be used to process low-resolution video image sequences and can be applied to the monitoring of people's walking status in larger physical spaces. It can also be used in nursing homes for medical care to assist in the prediction of elderly diseases and provide real-time early warning information to medical staff.
[0084] The present invention also discloses a medical gait recognition and prediction system based on a sequence model, comprising a camera, a gait video image acquisition module, a gait image data preprocessing module, a gait image task foreground segmentation module, a gait recognition and prediction module and a result output module;
[0085] The camera is used to capture gait video images;
[0086] The gait video image acquisition module is used to acquire gait motion video data from different viewing angles captured by a camera to obtain a gait image sequence;
[0087] The gait image data preprocessing module is used to perform standardization processing on the images in the gait image sequence;
[0088] The foreground segmentation module of the gait image task is used to obtain a character outline image sequence through a foreground segmentation model;
[0089] The gait recognition prediction module is used for the gait recognition prediction model based on the convolutional neural network and the sequence model. The recognition model and the prediction model are trained respectively, the gait sequence type data is obtained through the recognition model, and the type of the gait sequence diagram is output through the prediction model;
[0090] The result output module is used to match the type of gait sequence diagram with the disease database to obtain the disease category and grade.
[0091] The camera used in this system can be a fixed camera in a building or a camera installed on a service robot, which can walk with the person who needs to be observed.
[0092] This system uses fixed cameras in buildings or following service robots to obtain videos of people walking in physical space and converts them into sequence images. It obtains predicted gait sequence diagrams through recognition models and prediction models, and then obtains the type and level of symptoms through gait matching disease database, which makes it easier for medical staff to make disease judgments.
[0093] The present invention also discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the medical staff gait recognition and prediction method based on the sequence model is implemented.
[0094] In addition, the computer-readable storage medium of this embodiment may adopt any combination of one or more computer-readable storage media, wherein the computer-readable storage medium includes electrical, optical, electromagnetic, infrared or semiconductor systems, devices or components, or any combination thereof.
[0095] The present invention also discloses a computer device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; wherein:
[0096] The memory is used to store computer programs;
[0097] The processor is used to execute the steps of the medical gait recognition and prediction method based on the sequence model by running the program stored in the memory.
[0098] As an embodiment of the present invention, the communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc.
[0099] As an implementation mode of the present invention, the communication interface is used for communication between the above-mentioned terminal and other devices.
[0100] As an embodiment of the present invention, the memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.
[0101] As an embodiment of the present invention, the above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.
[0102] Different from the existing technology, this method uses a fixed camera in a building or a following service robot to obtain a video of a person walking in a physical space and convert it into a sequence of images. The segmentation model is used to identify and classify the abnormal gait of the research subject, and the Mamba sequence model is introduced to predict the gait, and based on this, the possible associated diseases are predicted. As a result, this method can be used to process low-resolution video image sequences and can be applied to monitoring the walking status of people in larger physical spaces. It can also be used in nursing homes for medical care to assist in the prediction of the elderly's diseases and provide real-time early warning information to medical staff.
[0103] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A medical gait recognition and prediction method based on a sequence model, characterized in that: The steps include: Gait video image acquisition steps: collecting gait motion video data from different perspectives to obtain a gait image sequence; Gait image data preprocessing steps: standardize the images in the gait image sequence; Foreground segmentation step of gait image task: obtain the character outline image sequence through the foreground segmentation model; Gait recognition prediction steps: Based on the gait recognition prediction model of convolutional neural network and sequence model, the recognition model and the prediction model are trained respectively, the gait sequence type data is obtained through the recognition model, and the type of gait sequence diagram is output through the prediction model.
2. A method for medical gait recognition and prediction based on a sequence model as claimed in claim 1, characterized in that: The gait video image acquisition step specifically includes acquiring a video image and performing frame processing on the video image at a predetermined threshold interval to acquire a gait image sequence G = {g1, g2, ...g n }, where G represents the gait image sequence, g i Represents one frame of a gait image in a gait image sequence.
3. A method for medical gait recognition and prediction based on a sequence model as claimed in claim 1, characterized in that: In the gait image data preprocessing step, the images in the gait image sequence are standardized, specifically: Standardize the input video images and remove images where the proportion of people is too large or too small; Subtract the redundant environment space that does not contain the person from the remaining image after culling; Resizes an image to a defined uniform pixel resolution and uniform size.
4. The method for medical gait recognition and prediction based on sequence model as claimed in claim 1, characterized in that: The foreground segmentation step of the gait image task is specifically as follows: The standardized images are divided into a training set, a test set and a validation set and input into the image segmentation model for image segmentation model training; The trained model detects the video image, outputs labels and masks in sequence, traverses the detection results, converts the instance segmentation mask into a binary contour image, and refines the binary contour image.
5. The method for medical gait recognition and prediction based on sequence model as claimed in claim 1, characterized in that: In the gait recognition prediction step, the recognition model is trained, specifically: the binary contour image sequence is input into the YOYO detection model training, and a plurality of input gait category sequences I are output at the same time, where I = {x1, x2, ..., x n1 }, x1 is the gait category.
6. A method for medical gait recognition and prediction based on a sequence model as claimed in claim 5, characterized in that: In the gait recognition prediction step, the prediction model is trained, specifically: the gait category sequence I is input into the Mamba model for prediction model training, and the predicted gait category sequence T is output, T = {z1, z2, ..., z n2 }, and n1>>n2, and n2≥1.
7. A medical gait recognition and prediction system based on a sequence model, characterized in that: It includes a camera, a gait video image acquisition module, a gait image data preprocessing module, a gait image task foreground segmentation module, a gait recognition prediction module and a result output module; The camera is used to capture gait video images; The gait video image acquisition module is used to acquire gait motion video data from different viewing angles captured by a camera to obtain a gait image sequence; The gait image data preprocessing module is used to perform standardization processing on the images in the gait image sequence; The foreground segmentation module of the gait image task is used to obtain a character outline image sequence through a foreground segmentation model; The gait recognition prediction module is used for the gait recognition prediction model based on the convolutional neural network and the sequence model. The recognition model and the prediction model are trained respectively, the gait sequence type data is obtained through the recognition model, and the type of the gait sequence diagram is output through the prediction model; The result output module is used to match the type of gait sequence diagram with the disease database to obtain the disease category and grade.
8. A medical gait recognition and prediction system based on a sequence model as claimed in claim 7, characterized in that: The result output module is used to match the type of gait sequence diagram with the disease database to obtain the disease category and level, specifically: Establishing a disease data set, wherein the disease data set includes gait data and levels corresponding to different diseases; Match the predicted gait category sequence T with the disease dataset; Output the disease with the highest matching probability and its level.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the medical gait recognition and prediction method based on a sequence model described in any one of claims 1 to 6 are implemented.
10. A computer device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; wherein: The memory is used to store computer programs; The processor is used to execute the steps of the medical gait recognition and prediction method based on the sequence model according to any one of claims 1 to 6 by running the program stored in the memory.