Motion Pattern Recognition Research Articles

Comparison of machine learning methods in sEMG signal processing for shoulder motion recognition

Machine learning (ML) methods have been previously applied and compared in pattern recognition of hand and elbow motions based on surface electromyographic (sEMG) signals. However, there are only a few studies that have investigated the ML methods for shoulder motion pattern recognition. This study compared the efficiency of ML algorithms, including support vector machine (SVM), logistic regression (LR), and artificial neural network (ANN) in processing sEMG signals for shoulder motion pattern recognition. This study also investigated the the effects of sliding time window epoch on the recognition accuracy. Eighteen healthy subjects were recruited for this study, their EMG signals were collected from twelve muscles during performing activities of daily living (ADL) motions including drinking, pushing forward/pulling backward, and abduction/adduction. The 80 % of recoded sEMG datasets were used for model training to build the ML models and 20 % were used for model validation and determination of the accuracy of ML algorithms in motion pattern recognition. The influence of sliding time window sizes was studied for algorithm optimization. Statistical analysis was performed to determine the difference in the accuracy of ML methods. Results showed that there was a significant difference among the three machine learning methods and different sliding time window sizes. There was not a significant difference in overlapping time. The highest accuracy was 97.41 ± 1.8 % using the SVM method with a sliding time window of 270 ms. Machine learning techniques provided a quick approach for shoulder motion pattern recognition. The better classifier for pattern recognition of shoulder motion was SVM.

Study of the relevance of gender in the classification of hand gestures by electromyography-based recognition systems

Some factors such as gender, age, physical fitness, and manual dominance are relevant and can influence the recognition of movement patterns using electromyography (EMG). In this scenario, we present an EMG signal analysis for men and women to observe if there is any significant difference. Data from 10 men and 10 women were acquired during the execution of six hand gestures (wrist flexion, wrist extension, wrist flexion to left, wrist extension to right, supination, and pronation) using eight channels armband. Four EMG time-domain signal features were extracted and hand gestures were classified using linear and quadratic discriminant analysis (LDA, QDA), and k-nearest neighbors (KNN) algorithms. Data of the feature difference absolute standard deviation value (DASDV) and waveform length (WL) were analyzed based on polar and bar graphs. KNN with 1 nearest neighbor obtained the best results between the classifiers for both men and women. For statistical analyses, the Wilcoxon-Mann-Whitney and Tukey post hoc in Friedman tests were used. The results show that there is no significant difference between data from different genders. After analyzing the results obtained and extensive comparison with related works, it was concluded that, for the conditions where the electrodes are positioned equidistantly, evaluating all the muscular groups of a limb (armband format), there was no significant difference observed between the data from different genders. In addition, this allows us to conclude that EMG armband on the forearm can be a good option for robotic systems control without the need for prior gender adjustment.

Personal‐specific gait recognition based on latent orthogonal feature space

Exoskeleton has been applied in the field of medical rehabilitation and assistance. However, there are still some problems in the interaction between human and exoskeleton, such as time delay, the existence of certain constraints on the human body, and the movement in time is hard to follow. A human motion pattern recognition model based on the long short-term memory (LSTM) is proposed, which can recognise the state of the human body. Meanwhile, the orthogonalisation method is integrated to make personal-specific disentangling, and it can effectively improve the generalisation ability of different groups of people, so as to improve the effective follower ability of the exoskeleton. Compared with some other traditional methods, this model has better performance and stronger generalisation ability, which has certain significance in the field of exoskeleton algorithm.

Method of Analyzing and Managing Volleyball Action by Using Action Sensor of Mobile Device

With the development of electronic technology and sensor technology, more and more intelligent electronic devices integrate micro inertial sensors, which makes the research of human action recognition based on action sensing data have great application value. Data‐based action recognition is a new research direction in the field of pattern recognition, which is essentially a process of action data acquisition, feature extraction, feature extraction, and recognition, the process of classification and recognition. Inertial motion information includes acceleration and angular velocity information, which is ubiquitous in daily life. Compared with motion recognition based on visual information, it can more directly reflect the meaning of action. This study mainly discusses the method of analyzing and managing volleyball action by using the action sensor of mobile device. Based on the motion recognition algorithm of support vector machine, the motion recognition process of support vector machine is constructed. When the data terminal and gateway of volleyball players are not in the same LAN, the classification algorithm classifies the samples to be tested through the characteristic data, which directly affects the recognition results. In this paper, the support vector machine algorithm is selected as the data classification algorithm, and the calculation of the classification process is reduced by designing an appropriate kernel function. For multiclass problems, the hierarchical structure of directed acyclic graph is optimized to improve the recognition rate. We need to bind motion sensors to human joints. In order to realize real‐time recognition of human motion, mobile devices need to add windows to the motion capture data, that is, divide the data into a small sequence of specified length, and provide more application scenarios for the device. This method of embedding motion sensors into devices to read motion information is widely used, which provides a convenient data acquisition method for human motion pattern recognition based on motion information. The multiclassification support vector machine algorithm is used to train the classification algorithm model with action data. When the signal strength of the sensor is 90 t and the speed is 2.0 m/s and 0.5 m/s, the detection accuracy of the adaptive threshold is 93% and 95%, respectively. The results show that the SVM method based on hybrid kernel function can greatly improve the recognition accuracy of volleyball stroke, and the recognition time is short.

Multimodal Sensor Motion Intention Recognition Based on Three-Dimensional Convolutional Neural Network Algorithm.

With the development of microelectronic technology and computer systems, the research of motion intention recognition based on multimodal sensors has attracted the attention of the academic community. Deep learning and other nonlinear neural network models have a wide range of applications in big data sets. We propose a motion intention recognition algorithm based on multimodal long-term and short-term spatiotemporal feature fusion. We divide the target data into multiple segments and use a three-dimensional convolutional neural network to extract the short-term spatiotemporal features. The three types of features of the same segment are fused together and input into the LSTM network for time-series modeling to further fuse the features to obtain multimodal long-term spatiotemporal features with higher discrimination. According to the lower limb movement pattern recognition model, the minimum number of muscles and EMG signal characteristics required to accurately recognize the movement state of the lower limbs are determined. This minimizes the redundant calculation cost of the model and ensures the real-time output of the system results.

Finite Class Bayesian Inference System for Circle and Linear Walking Gait Event Recognition Using Inertial Measurement Units.

Accurate and fast human motion pattern recognition is the key to ensuring lower limb assistive devices' appropriate assistance. The research on human motion pattern recognition of lower limb assistive devices mainly focuses on sagittal gait. The motion pattern such as circular walking (CW) is asymmetric about the sagittal plane of the body. CW is common in daily living. However, the recognition algorithm of CW is rarely reported. Since lower limb assistive devices interact with humans, lacking the capability of recognizing CW is dangerous. Thus, to realize the accurate and fast recognition of CW, this article proposed a finite class Bayesian interference system (FC-BesIS). FC-BesIS is designed to recognize walking activities (linear walking and CW) and gait events (heel contact, load response, mid stance, terminal stance, pre-swing, initial swing, mid swing, and terminal swing). A finite class method which reduces the number of potential classes according to elimination rules before decision-making is introduced. Elimination rules are designed based on likelihood estimation and sensor information. The experiments show that walking activities and gait events can be accurately and fastly recognized by FC-BesIS. The experiments also show that the performance of FC-BesIS in mean recognition accuracy (MRA) and mean decision time (MDT) is improved compared with BesIS. The MRA of walking activities and gait events are 100% and 97.38%, respectively. The MDT of walking activities and gait events are 28.19 ms and 33.94 ms, respectively. Overall, FC-BesIS has been proved to be an accurate and fast recognition algorithm for human motion patterns using wearable sensors.

Shoulder muscle activation pattern recognition based on sEMG and machine learning algorithms

Background and ObjectiveSurface electromyography (sEMG) has been used for robotic rehabilitation engineering for volitional control of hand prostheses or elbow exoskeleton, however, using sEMG for volitional control of an upper limb exoskeleton has not been perfectly developed. The long-term goal of our study is to process shoulder muscle bio-electrical signals for rehabilitative robotic assistive device motion control. The purposes of this study included: 1) to test the feasibility of machine learning algorithms in shoulder motion pattern recognition using sEMG signals from shoulder and upper limb muscles, 2) to investigate the influence of motion speed, individual variability, EMG recording device, and the amount of EMG datasets on the shoulder motion pattern recognition accuracy. MethodsA novel convolutional neural network (CNN) structure was constructed to process EMG signals from 12 muscles for the pattern recognition of upper arm motions including resting, drinking, backward-forward motion, and abduction motion. The accuracy of the CNN models for pattern recognition under different motion speeds, among individuals, and by EMG recording devices was statistically analyzed using ANOVA, GLM Univariate analysis, and Chi-square tests. The influence of EMG dataset number used for CNN model training on recognition accuracy was studied by gradually increasing dataset number until the highest accuracy was obtained. ResultsResults showed that the accuracy of the normal speed CNN model in motion pattern recognition was 97.57% for normal speed motions and 97.07% for fast speed motions. The accuracy of the cross-subjects CNN model in motion pattern recognition was 79.64%. The accuracy of the cross-device CNN model in motion pattern recognition was 88.93% for normal speed motion and 80.87% for mixed speed. There was a statistical difference in pattern recognition accuracy between different CNN models. ConclusionThe EMG signals of shoulder and upper arm muscles from the upper limb motions can be processed using CNN algorithms to recognize the identical motions of the upper limb including drinking, forward/backward, abduction, and resting. A simple CNN model trained by EMG datasets of a designated motion speed accurately detected the motion patterns of the same motion speed, yielding the highest accuracy compared with other mixed CNN models for various speeds of motion pattern recognition. Increase of the number of EMG datasets for CNN model training improved the pattern recognition accuracy.

A Multi-sensor School Violence Detecting Method Based on Improved Relief-F and D-S Algorithms

School bullying is a common social problem, and school violence is considered to be the most harmful form of school bullying. Fortunately, with the development of movement sensors and pattern recognition techniques, it is possible to detect school violence with artificial intelligence. This paper proposes a school violence detecting method based on improved Relief-F and Dempster-Shafe (D-S) algorithms. Two movement sensors are fixed on the object’s waist and leg, respectively, to gather acceleration and gyro data. Altogether nine kinds of activities are gathered, including three kinds of school violence and six kinds of daily-life activities. After wavelet filtering, 39 time-domain features and 12 frequency-domain features are extracted. To reduce computational cost, this paper proposes an improved Relief-F algorithm which selects features according to classification contribution and correlation. By drawing boxplots of the selected features, the authors find that the frequency-domain energy of the y-axis of acceleration can distinguish jumping from other activities. Therefore, the authors build a two-layer classifier. The first layer is a decision tree which separates jumping from other activities, and the second layer is a Radial Basis Function (RBF) neutral network which classifies the remainder eight kinds of activities. Since the two movement sensors work independently, this paper proposes an improved D-S algorithm for decision layer fusion. The improved D-S algorithm designs a new probability distribution function on the evidence model and builds a new fusion rule, which solves the problem of fusion collision. According to the simulation results, the proposed method has increased the recognition accuracy compared with the authors’ previous work. 89.6% of school violence and 95.1% of daily-life activities were correctly recognized. The accuracy reached 93.6% and the precision reached 87.8%, which were 29.9% and 2.7% higher than the authors’ previous work, respectively.

Recognition of motion patterns using accelerometers for ataxic gait assessment

The recognition of motion patterns belongs to very important research areas related to neurology, rehabilitation, and robotics. It is based on modern sensor technologies and general mathematical methods, multidimensional signal processing, and machine learning. The present paper is devoted to the detection of features associated with accelerometric data acquired by 31 time-synchronized sensors located at different parts of the body. Experimental data sets were acquired from 25 individuals diagnosed as healthy controls and ataxic patients. The proposed method includes the application of the discrete Fourier transform for the estimation of the mean power in selected frequency bands and the use of these features for data segments classification. The study includes a comparison of results obtained from signals recorded at different positions. Evaluations are based on classification accuracy and cross-validation errors estimated by support vector machine, Bayesian, nearest neighbours (k-NN], and neural network (NN) methods. Results show that highest accuracies of 77.1%, 78.9%, 89.9%, 98.0%, and 98.5% were achieved by NN method for signals acquired from the sensors on the feet, legs, uplegs, shoulders, and head/spine, respectively, recorded in 201 signal segments. The entire study is based on observations in the clinical environment and suggests the importance of augmented reality to decisions and diagnosis in neurology.

Recognition of Foot-Ankle Movement Patterns in Long-Distance Runners With Different Experience Levels Using Support Vector Machines.

Running practice could generate musculoskeletal adaptations that modify the body mechanics and generate different biomechanical patterns for individuals with distinct levels of experience. Therefore, the aim of this study was to investigate whether foot-ankle kinetic and kinematic patterns can be used to discriminate different levels of experience in running practice of recreational runners using a machine learning approach. Seventy-eight long-distance runners (40.7 ± 7.0 years) were classified into less experienced (n = 24), moderately experienced (n = 23), or experienced (n = 31) runners using a fuzzy classification system, based on training frequency, volume, competitions and practice time. Three-dimensional kinematics of the foot-ankle and ground reaction forces (GRF) were acquired while the subjects ran on an instrumented treadmill at a self-selected speed (9.5–10.5 km/h). The foot-ankle kinematic and kinetic time series underwent a principal component analysis for data reduction, and combined with the discrete GRF variables to serve as inputs in a support vector machine (SVM), to determine if the groups could be distinguished between them in a one-vs.-all approach. The SVM models successfully classified all experience groups with significant crossvalidated accuracy rates and strong to very strong Matthew’s correlation coefficients, based on features from the input data. Overall, foot mechanics was different according to running experience level. The main distinguishing kinematic factors for the less experienced group were a greater dorsiflexion of the first metatarsophalangeal joint and a larger plantarflexion angles between the calcaneus and metatarsals, whereas the experienced runners displayed the opposite pattern for the same joints. As for the moderately experienced runners, although they were successfully classified, they did not present a visually identifiable running pattern, and seem to be an intermediate group between the less and more experienced runners. The results of this study have the potential to assist the development of training programs targeting improvement in performance and rehabilitation protocols for preventing injuries.

A novel method to process surface electromyography signal for pedestrian lower limb motion pattern recognition

Surface electromyography (sEMG) signals mainly contain power line interference (PLI), white Gaussian noise (WGN), and baseline wandering (BW) noise. These noises lead to the problems of poor feature extraction performance and low recognition rate. In this paper, we propose a novel sEMG signal processing method named filtering and self-enhancement algorithm with classical wavelet (FSECW) algorithm, which denoises interference noises of raw signals and improve the pedestrian motion pattern recognition rate. The proposed FSECW algorithm contains two core parts: in the first part, the original signal is reconstructed after four-layer wavelet decomposition. This part decreases the BW noise and enhances the active segment; in the other part, band-pass filtering and lifting wavelet transformation are used to reduce noises inside and outside the sEMG signal band. Then two signals from the above parts are multiplied. Thus, the enhanced filtered signal of the active segment is obtained. After feature extraction, the algorithm uses classical machine learning algorithm for motion pattern recognition. Experimental results show that the proposed FSECW algorithm does not need to set different thresholds for different data sets with the same motion pattern. Moreover, it has better adaptability to conversions of different motion patterns.

Effects of Different Feature Parameters of sEMG on Human Motion Pattern Recognition Using Multilayer Perceptrons and LSTM Neural Networks

In response to the need for an exoskeleton to quickly identify the wearer’s movement mode in the mixed control mode, this paper studies the impact of different feature parameters of the surface electromyography (sEMG) signal on the accuracy of human motion pattern recognition using multilayer perceptrons and long short-term memory (LSTM) neural networks. The sEMG signals are extracted from the seven common human motion patterns in daily life, and the time domain and frequency domain features are extracted to build a feature parameter dataset for training the classifier. Recognition of human lower extremity movement patterns based on multilayer perceptrons and the LSTM neural network were carried out, and the final recognition accuracy rates of different feature parameters and different classifier model parameters were compared in the process of establishing the dataset. The experimental results show that the best accuracy rate of human motion pattern recognition using multilayer perceptrons is 95.53%, and the best accuracy rate of human motion pattern recognition using the LSTM neural network is 96.57%.

Exploring the Application of Pattern Recognition and Machine Learning for Identifying Movement Phenotypes During Deep Squat and Hurdle Step Movements.

BackgroundMovement screens are increasingly used in sport and rehabilitation to evaluate movement competency. However, common screens are often evaluated using subjective visual detection of a priori prescribed discrete movement features (e.g., spine angle at maximum squat depth) and may not account for whole-body movement coordination, or associations between different discrete features.ObjectiveTo apply pattern recognition and machine learning techniques to identify whole-body movement pattern phenotypes during the performance of exemplar functional movement screening tasks; the deep squat and hurdle step. Additionally, we also aimed to compare how discrete kinematic measures, commonly used to score movement competency, differed between emergent groups identified via pattern recognition and machine learning.MethodsPrincipal component analysis (PCA) was applied to 3-dimensional (3D) trajectory data from participant’s deep squat (DS) and hurdle step performance, identifying emerging features that describe orthogonal modes of inter-trial variance in the data. A gaussian mixture model (GMM) was fit and used to cluster the principal component scores as an unsupervised machine learning approach to identify emergent movement phenotypes. Between group features were analyzed using a one-way ANOVA to determine if the objective classifications were significantly different from one another.ResultsThree clusters (i.e., phenotypes) emerged for the DS and right hurdle step (RHS) and 4 phenotypes emerged for the left hurdle step (LHS). Selected discrete points commonly used to score DS and hurdle step movements were different between emergent groups. In regard to the select discrete kinematic measures, 4 out of 5, 7 out of 7 and 4 out of 7, demonstrated a main effect (p < 0.05) between phenotypes for the DS, RHS, and LHS respectively.ConclusionFindings support that whole-body movement analysis, pattern recognition and machine learning techniques can objectively identify movement behavior phenotypes without the need to a priori prescribe movement features. However, we also highlight important considerations that can influence outcomes when using machine learning for this purpose.

Fingerprint-inspired graphene pressure sensor with wrinkled structure

Nowadays, flexible pressure sensors have broad applications in healthcare monitoring, wearable electronics and so on. However, conventional fabrications are mostly time-consuming, complicated and even use toxic chemical reagents. It is in urgent need to fabricate pressure sensors with high sensitivity in a facile way. In nature, the fingertip skin of human has many epidermal ridges, which can sense small external stimuli. Here, inspired from the fingerprint, we propose a new type of pressure sensor made with reduced graphene oxide (rGO). The rGO film is designed to form a wrinkled structure by a facile stretching-releasing process, which mimics the morphology of fingertip skin. Owing to the fingertip-skin-like wrinkled structure, the pressure sensor exhibits high sensitivity of 5.77 kPa−1 (0 - 490 Pa). It also has a lot of attractive properties, such as fast response (< 100 ms), small detecting pressure (3 Pa) and good repeatability. We show the ability of the pressure sensor in small pressure detecting, human motion monitoring and motion pattern recognition. Furthermore, a translation system and a multi-touch device with spatially resolved detection based on the pressure sensor are developed. This pressure sensor is hopeful to have potential applications in wearable electronics and human–machine interaction systems.

Evaluation of Professional Ballet Dancers Body Posture During Barre Movements.

Ballet dancers have finite careers due to the demands placed upon their bodies throughout years of training, study, and performance. The average age a dancer retires is 34 due to persistent stress on their joints and injuries due to mechanical overload or overuse. Proper form is crucial to prevent injury. The aim of this study was to establish benchmarks for normal movement patterns among professional dancers. Ten professional ballerinas were studied. Reflective markers were placed on the pelvis, left and right anterior superior iliac spine (ASIS), and posterior iliac spine (PSIS) to evaluate motion during Barre movements: plié, grand battement, and développé. Pelvis flexion/extension, mediolateral rotation, and torsion were analyzed. These motions test different skills. The plié is a controlled coordinated motion using both legs. The grand battement and développé both require leg extension, one with a quick motion that creates momentum and one using controlled motion that requires strength. Each requires core and pelvis stability to perform accurately and with less injury. Dancers' motions were consistent. Maximum pelvis range of motion for the plié, grand battement, and développé were 8.0, 42, and 50 deg, respectively. This represents usable benchmarks with which other dancers may be compared, for example, those who are at different levels of training, injured, predisposed to injury, or recovering from injury. Early recognition of pathologic movement patterns could benefit professional and amateur dancers by helping to prevent injuries, and potentially improve the quality and length of their careers.

Intelligent human computer interaction based on non redundant EMG signal

Human computer interaction plays an increasingly important role in our life. People need more intelligent, concise and efficient human-computer interaction. It is of great significance to optimize the process of human-computer interaction by using appropriate calculation methods. In order to eliminate the interference data of thumb recognition based on sEMG signal in the process of human-computer interaction, simplify the data processing, and improve the working efficiency of general equipment of sEMG signal. In the process of gesture recognition using sEMG signals generated by thumb, a method of redundant electrode determination based on variance theory is proposed. The redundancy of five groups of action signals is divided into 16 levels and visualized. By comparing the results of thumb motion recognition when different redundant channels are removed, the optimal channel combination in the process of thumb motion recognition is obtained. Finally, two kinds of classifiers suitable for sEMG signal field are selected, and the classification results are compared, and the best method of thumb motion pattern recognition is obtained.

Human Body Mixed Motion Pattern Recognition Method Based on Multi-Source Feature Parameter Fusion.

Aiming at the requirement of rapid recognition of the wearer’s gait stage in the process of intelligent hybrid control of an exoskeleton, this paper studies the human body mixed motion pattern recognition technology based on multi-source feature parameters. We obtain information on human lower extremity acceleration and plantar analyze the relationship between these parameters and gait cycle studying the motion state recognition method based on feature evaluation and neural network. Based on the actual requirements of exoskeleton per use, 15 common gait patterns were determined. Using this, the studies were carried out on the time domain, frequency domain, and energy feature extraction of multi-source lower extremity motion information. The distance-based feature screening method was used to extract the optimal features. Finally, based on the multi-layer BP (back propagation) neural network, a nonlinear mapping model between feature quantity and motion state was established. The experimental results showed that the recognition accuracy in single motion mode can reach up to 98.28%, while the recognition accuracy of the two groups of experiments in mixed motion mode was found to be 92.7% and 97.4%, respectively. The feasibility and effectiveness of the model were verified.

Spatial movement pattern recognition in soccer based on relative player movements.

Knowledge of spatial movement patterns in soccer occurring on a regular basis can give a soccer coach, analyst or reporter insights in the playing style or tactics of a group of players or team. Furthermore, it can support a coach to better prepare for a soccer match by analysing (trained) movement patterns of both his own as well as opponent players. We explore the use of the Qualitative Trajectory Calculus (QTC), a spatiotemporal qualitative calculus describing the relative movement between objects, for spatial movement pattern recognition of players movements in soccer. The proposed method allows for the recognition of spatial movement patterns that occur on different parts of the field and/or at different spatial scales. Furthermore, the Levenshtein distance metric supports the recognition of similar movements that occur at different speeds and enables the comparison of movements that have different temporal lengths. We first present the basics of the calculus, and subsequently illustrate its applicability with a real soccer case. To that end, we present a situation where a user chooses the movements of two players during 20 seconds of a real soccer match of a 2016–2017 professional soccer competition as a reference fragment. Following a pattern matching procedure, we describe all other fragments with QTC and calculate their distance with the QTC representation of the reference fragment. The top-k most similar fragments of the same match are presented and validated by means of a duo-trio test. The analyses show the potential of QTC for spatial movement pattern recognition in soccer.

Movement Pattern Recognition in Physical Rehabilitation - Cognitive Motivation-based IT Method and Algorithms

In this paper, a solution is presented to support both existing and future movement rehabilitation applications. The presented method combines the advantages of human-computer interaction-based movement therapy, with the cognitive property of intelligent decision-making systems. With this solution, therapy could be fully adapted to the needs of the patients and conditions while maintaining a sense of success in them, thereby motivating them. In our modern digital age, the development of HCI interfaces walks together with the growth of users’ needs. The available technologies have limitations, which can reduce the effectiveness of modern input devices, such as the Kinect sensor or any other similar sensors. In this article, multiple newly developed and modified methods are introduced with the aim to overcome these limitations. These methods can fully adapt the movement pattern recognition to the users' skills. The main goals are to apply this method in movement rehabilitation, where the supervisor, a therapist can personalize the rehabilitation exercises due to the Distance Vector-based Gesture Recognition (DVGR), Reference Distance-based Synchronous/Asynchronous Movement Recognition (RDSMR/RDAMR) and the Real-Time Adaptive Movement Pattern Classification (RAMPC) methods.

Research on motion pattern recognition of exoskeleton robot based on multimodal machine learning model

Exoskeleton as a real-time interaction with the wearer’s intelligent robot, in recent years, becomes a hot topic mouth class research in the field of robotics. Wearable exoskeleton outside the body, combined with the organic body, plays a role in the protection and support. By wearing an exoskeleton robot, it is possible to expand the wearer’s athletic ability, increase muscle endurance, and enable the wearer to complete tasks that he or she cannot perform under natural conditions. Based on the above advantages, the exoskeleton robot in military medical care and rehabilitation has broad application prospects. This paper describes the multimodal model of machine learning research status and research significance of the text on the exoskeleton robot applications, and on the basis of a detailed study of gait. It mainly involves: analysis and planning and obtaining motion information processing, pattern recognition and analysis of gait and the gait conversion process, and the EEG and joint position, foot pressure, such as different modes of data as input to machine learning models to improve the timeliness, accuracy and safety of gait planning. Since the common movement process involves the transformation process of gait, this paper studies the gait transformation process including squatting, walking on the ground and standing.