Research Of Human Skeleton Based Behavior Understanding And Application In Remote Human-robot Interaction System

Human skeleton extraction and behavior understanding have important research value and application prospects in the fields of human-robot interaction,artificial intelligence and rehabilitation medicine.The human skeleton can effectively describe the posture and movement of the human body and provide powerful support for human behavior understanding.However,due to the different shapes of the human body and the varied movements,self-occlusion problems may occur and the extraction of the human skeleton may be difficult.In addition,how to effectively extract features from the human body during exercise is also a difficult point to understand.In this paper,aiming at the above problems,the method of human skeleton extraction based on model fitting is proposed and the human skeleton is used to realize the understanding of human behavior.And it is applied in remote human-robot interaction system and rehabilitation training.For the self-occlusion problem caused by human body motion changes in human skeleton extraction,and the traditional 2D thinning algorithm cannot effectively solve this problem and will result in the missing of human skeleton curve,this paper proposes an adaptive segmentation algorithm and multi-layer 2.5D thinning algorithm.The algorithm combines the depth image segmentation and thinning algorithm to make up for the lack of detection of the traditional thinning algorithm.It effectively solves the problem of extracting the human body skeleton curve from the occlusion region and improves the robustness of the detection results.For the problem that the human posture is diverse and difficult to extract the human skeleton,a standard skeleton model based on a tree structure is defined in this paper,and the interaction between each skeleton joint is represented by a kinematic chain.At the same time,this paper defines eight kinds of posture matching templates,and combines the distance lines outside and inside the thin line to fit the reference points.Finally,the standard skeleton model is fitted to the image and the complete human skeleton is detected.In addition,in order to solve the problem that human behavior is difficult to analyze and understand,this paper builds a behavioral understanding model based on skeleton information and combined with a recurrent neural network to achieve the encoding and feature extraction of skeleton motion sequences.In order to verify the accuracy of the algorithm,this paper compares and analyzes the human skeleton extraction experiments on the Stanford database and the UESTC-SE database collected and labeled on its own,and proves that the algorithm can achieve 90% detection accuracy and 0.023 frames/second.The real-time processing speed is 10 times faster than similar algorithms.At the same time,the behavior understanding experiment was performed on the MSRDaily3 D database,and 87% recognition rate was obtained to verify the effectiveness of the algorithm.This paper designs and develops a remote human-machine interaction system based on a virtual robot,which uses a virtual robot as an interactive visual medium.We combined this system with the lower extremity rehabilitation exoskeleton robot.Through the extraction of the human skeleton and the understanding of human behavior,a remote rehabilitation training model was implemented to improve rehabilitation and the efficiency of the tour also increases the interest of the training process.