Research And Application Of Human Motion Segmentation Methods

Video surveillance technique is one of the new science research aspects which have huge development potential,and it is applied more and more in subway industry. Detecting of human motion is one of the key techniques of intelligent video surveillance.Human motion detection includes motion segmentation and object classification.Motion segmentation is an important part of human motion detection. This thesis is based on the project"the application of embedded technology research in mass transit surveillance system(project number:06DZ15003)",it starts from the need of detection of an unattended-package.First of all,the paper discusses and analyzes human motion segmentation technologies,and presents corresponding improvement on the basis of analysis the strengths and weaknesses of segmentation algorithm.And then builds a prototype system of human motion segmentation after studying each improved algorithm.Finally the paper realizes the functions of object segmentation,object tracking and behavior recognition.Combined with the requirements of unattended packages detection system under the rail transportation environment,a method of human motion segmentation based on video is proposed.Frame-difference method,threshold segmentation,morphologic filter and image filling are included,and the corresponding solutions for the problems and difficulties in the implementation are proposed.With regard to human motion segmentation,the basic principles of frame-difference method are analyzed.An improved frame-difference method was proposed which separates three frames from the video image for human motion segmentation.Moreover,human motion segmentation under different illumination is analyzed. An improved RGB subtraction method which is used to segment moving targets and get the relevant information is proposed.And the experimental results are presented.In view of real-time performance of the system,an improved pixel of frame traversing method is proposed.The experimental results show the effection to improve the real time performance of the systemFinally,a prototype system of unattended-package detection and recognition is implemented based on VC++ and OpenCV.