Research On Pedestrian Detection And Tracking Based On Video

Pedestrian detection and tracking has been a research topic of computer vision area due to its widely application, such as Intelligent Video Surveillance, Human computer interface, virtual reality, statistics of visitor flow-rate, etc. However, human body is a non-rigid object, which results in widely difference around different human body, also, some background interferences such as trees, power lines, tyres are similar with human body. All of reasons above make the work of pedestrian detection and tracking more difficult. Features and methods adopted by pedestrian detection and tracking based on statistical learning method are mainly researched and analyzed, especially on pedestrian detection in Intelligence Video Surveillance area, the main content of this paper are as followings:1. The research background, status of pedestrian detection and tracking based on video and the topic detection and tracking method in recent years are respectively introduced, which lay a solid theoretical foundation for future pedestrian detection and tracking.2. The characteristics of the pedestrian object are analyzed, combined with the Intelligent Video Surveillance application, a marking foreground method is proposed that the foreground is extracted and marked, so that only the foreground pixels are detected instead of all of the image at the detection stage, which can improve the accuracy of detection, and make the detection speed more faster.3. The pedestrian detection methods based on statistical methods are analyzed and combined with the characteristics of cascade classifier-the more of the cascade classifier lever, the higher of the False Negative, a method of merged cascade classifier is proposed, which can improve the detection rate on the condition that the performance will not decrease.4. The Mean Shift algorithm is applied on the pedestrian tracking area, based on the insufficient of Mean Shift algorithm, this paper proposed a new object model updated method, which makes Mean Shift algorithm more robust to the background and human body’s physical changes.5. Based on the study and analysis above, this paper completes a rapid and more accurate pedestrian detection system based on merged classifiers and a pedestrian tracking system based on improved Mean Shift algorithm, and the experiment results at the standard available pedestrian database verity the effectiveness and the practical applicability of the algorithm proposed in this paper.