Research And Application Of The Key Technologies In 3D Human Motion Analysis For Human Computer Interaction

Human motion analysis is an important research issue in human computer interation（HCI）,accurate and fast identification of human motions is the premise and guarantee of real-time interaction in the human-centered HCI system.Most of the early human motion analyses were sequenced based on two-dimensional video image,which is still a challenge and its application is limited for two-dimensional image is apt to be affected by integrated factors such as illumination variation,complicated background,camera perspective.In recent years,with the rapid development of depth transducer technology,3D human skeleton sequence can be collected accurately with the help of cheap depth transducer,because that 3D human skeleton sequence not only can describe human motion more intuitively and precisely,but avoid some problems of two-dimensional video image.As a result,the human motion analytical research is greatly promoted.This paper carried out motion analytical research based on 3D human skeleton sequence.As there are quantities of redundant frames in motion posture sequence due to the high-frequency sampling of sensor device,research in this paper mainly focuses on feature description of posture,key-pose frame extraction of motion,motion recognition and application,including:（1）.3D skeleton sequence acquisition of human motion and feature description of motion and posture.Human skeleton articulation data could be collected by depth sensor to build human coordinate system,and the static posture in motion sequence could be described by relative distance between characteristic articulation points.Then the motion data set Visswust₃DMotion could be built based on the actual project,as well as the main body motion in application system of man-machine interaction,to lay a foundation for subsequent human motion analysis.（2）.The research about key-pose frame extraction technology of human motion.Firstly,the current available clustering methods in literatures are difficult to determine the number of key pose frames automatically,and may destroy the postures’ temporal relationships while extracting key frames.To deal with this problem,this paper proposes a new key pose frames extraction mehod on the baiss of 3D space distances of joint points and the time-constraints X-Means algorithm.The experimental results show that this method can automatically determine the number of key frames and save the temporal charactersitics of motion frames according to the motion pose sequence.Secondly,for the questions that complex error threshold adjustment in current curve simplified method to extract key frame and the extracted key frames cannot meet the needs of users,a 3D motion key frame extraction method based on interaction and visualization is proposed,and in accordance with this,a set of 3D human motion generation system is designed.The laboratory shows that this method makes users obtain proper motion key frame sequence and generate the standard motion library more conveniently,more accurately,and more efficiently.（3）.Technical research of human motion recognition.Human motion recognition method based on DTW and CHMM were realized respectively on the basis of analysis on current major human motion recognition methods.Additionally,human motion recognition method based on key posture and DTW was proposed for recognition method based on original posture sequence and DTW were characterized by low recognition rate,slow recognition speed and weak robustness.The experimental result showed that this method had higher recognition rate,faster recognition speed and better robustness than DTW recognition method based on the original posture sequence.Although its recognition rate was a little lower than CHMM recognition method,this method had greater advantages on algorithm complexity and recognition speed.Therefore this paper recommended the human motion recognition method based on key posture and DTW matching to be used in man-machine interaction system.（4）.A motion training system of virtual traffic police is designed and developed based on Unity3 D platform and the human motion recognition method proposed in this paper,so as to solve problems such as the boring training process,excessive dependence on on-site conducting experience of traffic police and difficulty in recording training data during the motion conducting training of traffic police.The experimental result showed that this system had certain application values in conducting motion teaching of traffic police and intelligent transportation field.