Research And Application Of Real-time Target Position And Attitude Recognition And Motion Tracking

With the advent of robots in the 1950 s,robots have been widely used in various fields.The industrial field is an important application scene of robots whose heavy workload and low-security working environment make labor costs high.Therefore,the safety,real-time and programmability characteristics of industrial robots are the key to solving the above problem.Gesture recognition is flexible and intuitive,which brings new ideas to remote human-computer interaction.In recent years,it has become a research hotspot of human-computer interaction.Summarizing the current research on robot control and gesture recognition,the following background points are obtained:(1)In actual scenarios,the operator cannot leave the scope of device wiring for robots using hand-held terminals for control.So the ease and security of human-computer interaction still can be improved;(2)Most of the research on gesture recognition focuses on artificial neural networks and hidden Markov models,and there are few studies on geometric features;(3)The application research of gesture recognition in robot remote control is still rare.This thesis is aimed at the above points.Firstly,the thesis studies the theory of several geometric feature recognition methods.Considering the stability and real-time,a more mature gesture recognition method based on fingertip detection is selected.Then based on the boundary features of gestures,the principle of the convex hull algorithm and the realization method of the traditional convex hull algorithm are studied.On this basis,an improved fast convex hull algorithm is proposed.Finally,a gesture recognition is proposed based on the fast convex hull algorithm which defines the geometric characteristics of gestures.The thesis still designs a set of gesture instructions to apply it to the robot's remote human-computer interaction for experimental verification.The main contents of this thesis are as follows:1.Studied the principle of Kinect's depth measurement and bone tracking.Analyzing the connection between actual distance and offset distance to obtain the error range of Kinect depth measurement.Studied the principle of the threshold segmentation algorithm and proposed a depth-based threshold segmentation method.Then compared with the Otsu threshold segmentation algorithm.Systematically summarized the principles and ideas of traditional median filter algorithm and fast value filter algorithm,and compared the efficiency of the three median filter algorithms through experiments.Analyzed the characteristics of different gesture recognition technologies,and clarified the reason for using gesture recognition methods based on fingertip detection in this thesis.2.Analyzed and researched the principle of convex hull algorithm,and summarized the algorithm ideas of the traditional Graham scanning method and Jarvis stepping method.For the above algorithms,this thesis proposed an improved fast convex hull algorithm: First,the point set is divided into two subsets by selecting the appropriate dividing line.Second,connecting the points farthest from the upper and lower subsets of the dividing line to form a triangle,and rounding off the points inside the triangle and the non-vertex points on the line.Finally,repeating the above operations for the subset outside the triangle until the subset is empty,and the remaining points are the convex hull points.Then the thesis verifies the effectiveness of the improved algorithm through experiments,and compares with the traditional convex hull algorithm to prove the realtime performance of the improved algorithm.3.The thesis proposes a gesture recognition algorithm based on convex hull set.For two different types of gestures,the algorithm is divided into two parts: digital gesture recognition and angle gesture recognition.The former uses a convex hull set to generate a polygonal fitting curve,and compares it with the contour of the gesture to find the concave set.The gesture is determined by the geometric relationship between the center point of the hand,the center of the convex hull,the concave set and the convex hull set.The latter is based on the mathematical change of the connection between specific points in the convex hull set in the two-frame interval image to determine the gesture.Through experiments,the average recognition rate of different gestures is 95%~100%,and the trajectory recognition rate is more than 97%.4.Focusing on the application of gesture recognition in human-computer interaction,and using the gesture recognition algorithm proposed in this thesis,this thesis designed a robotic arm control system to verify the above results.Through the changes of gestures and palm trajectories,the operator can lead the 4,5,and 6 axes of the robot arm to perform displacement,roll,rotation and pitch.The average synchronization rate is above 98%.The research work in the thesis has certain reference value for the application of gesture recognition in robot remote control.