Research On Moving Object Detection And Tracking Algorithms Based On Video Image

Video surveillance system(VSS) is the main trend of modern defence system because of its visibility, convenience and abundance in content and has been widely used in many fields where a real-time surveillance is needed, such as national defense, traffic control, the intelligent public security and so on. Nowadays, VSS still depends on manual operation, which wastes resources and affects the efficiency, so studying the typical algorithms used in video surveillance and designing an intelligent video surveillance system is very important.Moving object detection and tracking are important parts of video monitor system and play important roles to other topics' progress in computer vision. So how to detect and track object steadily, real-timely and effectively, becomes an important problem that needs to be paid attention and researched. The paper studied the key technologies of the field based on the current research achievements and mainly studied about the technologies on human object detection, segmentation and tracking.During object detection, several object foreground detection algorithms widely used are introduced, and their performances are analyzed. During object localization and segmentation, due to head contours similar to circles seen from the video sequences captured from vertical angular camera, the paper combines Freeman chain code and Random Sample Consensus (RANSAC) algorithm to estimate circle parameters and then localizes the human object quickly and accurately.During object tracking, in the light of current Mean Shift tracking algorithm's drawbacks, Kalman filter and Mean Shift tracking algorithm integrating with color, texture and motion information is proposed and its tracking performance has been improved than before. Aiming at the posture's significant change of a single object, a novel tracking algorthim based on RGB histogram feature, LBP (Local Binary Pattern) histogram feature and PPBTF (Pixel-Pattern-Based Texture Feature) histogram feature, using the semi-supervised learning is proposed on the basis of machine learning theory.