The Research And Application Of Human Motion Detection And Tracking Techniques

The detection and tracking of human motion is not only an important part of motion analysis and behavior identification,but also a research emphasis of computer vision,it involves a wide range application foreground and great economic value in smart monitoring system,advanced user interface and analysis of human movement.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 and tracking of an unattended-package.First of all,the paper discusses and analyzes of human motion detection and tracking technologies, and presents corresponding improvement on the basis of analyzing the strengths and weaknesses of detection and tracking algorithm,and then builds a prototype system of human motion detection and tracking after studying the experimental results of improved algorithm.Finally the paper realizes the functions of object detection,object tracking and behavior recognition.The main works done in this paper are as follows:(1)According to the issues of complicated background,illumination changes, shadows,and so on,the paper has done researches in establishing background using Gaussian model.We can change the fixed background into a changing background that according to the environment by setting the update parameters.This not only makes the detection of object more accurate but also reduces the probability of mistaking foreground for background.(2)When using frame differencing method to detect motion object,for the incompletion of objects,the paper adopts frame differencing method based on OpenCV to extract moving targets so that we can maintain the integrity of the object. In addition,after specifically analyzing the advantages and disadvantages of background subtraction and optical flow,a new method combined both is put forward. (3)Combining the application of unattended-package,when detecting,tracking and recognizing packages and human bodies,we try to adopting the template matching method for detection on one side.We can detect the object accurately according to the match percent,at the same time the paper presents multi-template matching method for real-time tracking.The using of multi-template method can omit the segment of template updating.On the other side,we use the classical Kalman filter for the position prediction and updating on the aspect of real-time tracking.(4)This paper implements a prototype system based on OpenCV,the modules of videos reading,targets detection,objects tracking and locating and behaviors recognition are included.