Research On Posture Recognition And Robot Tracking Control System

With the rapid development of intelligent technology,robots have begun to enter everyone's work and life.In industry,robots can replace workers in dangerous jobs to protect their safety,and they can also perform repetitive and boring actions to reduce the burden on workers.In daily life,robots play more and more roles,such as sweeping robots,bowling robots,dancing robots,piano robots,etc.The application of robots is inseparable from human control,so how to carry out simple and effective humancomputer interaction control is very important.There are many ways of humancomputer interaction.We can choose posture information as input,which does not rely on intermediary items such as mouse and keyboard.In this paper,we use posture information as input to control the robot to achieve tracking control functions.This paper focuses on the posture recognition algorithm.First,use WTS(weighted tangent segmentation algorithm)to complete the segmentation of complex postures,and then use SOM(self-organizing map)neural network based on SSED(sum of squared errors distribution)to cluster the segmented simple postures to obtain the basic posture and Basic posture data set.Finally,the BP(back propagation)neural network is used to complete the recognition of simple postures,and the recognition results of complex postures are obtained according to the timestamps obtained by segmentation.In terms of robot tracking control,this paper only does a preliminary study.We designed the posture recognition system and the robot control system respectively,and realized the tracking control of the robot according to the received recognition data.This article conducts research from the following aspects:1.The commonly used coordinate systems in three-dimensional space are introduced,and the posture data of MPU9250 are described and analyzed.Through the analysis of the human body structure and the basic range of motion of the human body,a simplified human body model is obtained,and a humanoid robot model is built according to the simplified human body model.2.This paper propostures a weighted tangent segmentation algorithm to complete the segmentation of complex postures.The complex posture is compostured of multiple simple postures.Between the two simple postures,there must be an area with stable data.In this area,the movement trend of the previous simple posture gradually weakens,and the movement trend of the next simple posture gradually begins.According to this stable area,separate different simple postures.3.This paper proposes a SOM neural network clustering algorithm based on SSED to complete simple posture clustering.For different numbers of clusters,we use SSE to determine the clustering effect,so as to obtain the best number of clusters,realize simple posture clustering and get the basic posture data set.4.This paper propostures a BP neural network algorithm based on timestamp to recognize complex postures.The segmentation algorithm divides complex postures to obtain simple postures and their timestamp sequences.The BP neural network is used to classify and recognize each simple posture on the time stamp,and obtain the recognition sequence of the complex posture.5.In this paper,we designed the posture recognition system and the robot tracking control system respectively.In the posture recognition system,it mainly includes three modules: data acquisition,main controller and wireless transmission,which complete the acquisition,processing,recognition and transmission of posture data.The robot tracking control system mainly includes three modules: the main controller,wireless transmission and steering gear control board,which completes the reception of posture data and the tracking control of the robot.