The Application Research Of Machine Learning In Gait Recognition For Human Exo-skeleton System

Exoskeleton system is a highly human-machine coupled systems, and it combineshuman intelligence and mechanical strength，which is a mechanical structure integratesadvanced control, information and communications technology. Machine Learning isused to complete gait analysis for flexible control.The prototype is used to support the payload of human body in walking, goingupstairs or downstairs and so on. This paper focuses on the perceptive subsystem, whichis combined to control subsystem to enhance the pilot’s strength and realized assistingthe wearer’s walking. Based on foot pressure sensors and the rotary encoders on knee,the gait can be recognized. In control system, the position control loop is built. On thisbasis, the machine learning classification algorithms and clustering algorithms are usedfor offline data mining on a data collection of prototype, in order to establish a classifierfor movement prediction and judgment of lower limb. At the same time, the articleanalyses the procedure of standing from sitting by clustering for rehabilitationexoskeleton. Experiments mainly refer to The C4.5decision tree algorithm, clusteringalgorithm k-means and association rules. Software simulation results show that theclassifier can make use of gait analysis data accurately to judge the action, and thecluster can cluster lower limb motion. Due to machine learning are very muchdependent on the sample data, the training sample is good or bad directly affect themodel’s excellent and bad, which affect the accuracy of gesture recognition. So inaccordance with the need of human gait analysis in reference to the standard on thebasis of normal gait, and according to the prototype design intent and control status quo,the suitable model of a system gait was established. In view of the initial sensor data,data mining technology, those are attribute select analysis and anomaly detection, wereused to optimize the overall characteristics of the training sample data.Simulation results show that the test data could be accurately classified with machinelearning. Effectively using of these results can offer great convenience to the flexibilitycontrol for exoskeleton so that exoskeleton system could achieve a high degree ofhuman-computer coupling.