Research On Moving Object Detection And Tracking And System Implementation

Research on moving object detection and tracking are important areas of com-puter vision. In this thesis, we first analyze and compare algorithms and models onthe areas. Then we propose an efficient moving object detecting and tracking systemwhich performs better than existing systems in detecting moving objects from dynamicscene. The system is based on phase discrepancy motion detection algorithm. We us-ing Canny-like salient area extractor, image segmentation and object tracking in thesystem to generate excellent results.The main contributions of the thesis include the following:1. We analyze phase discrepancy motion detection algorithm in detail and pointout the ?aws of the method in applying to a practical system through theoretical analy-sis and experiment results. We propose our solution according to the ?aws. We designand implement a Canny-like salient area extractor to extract salient areas from motionsaliency map. We use the result of image segmentation to connect different salientareas belonging to the same object and extend salient areas to whole moving object.We then get more accurate bounding boxes through the properties of adjacent motionsaliency maps. Experiments show that our system is good at detecting moving objectsin dynamic scene and suitable for a practical motion detection system.2. Mean shift object tracking algorithm and particle filter object tracking algo-rithm are analyzed and compared. We point out the possibility of incorporate theminto moving object detecting system based on the analysis. We use object trackingmethod in our system for two purpose: to enhance the stability of detecting results andto track interesting objects.3. Many techniques, such as multi-threading, are applied during the implemen-tation of the demo system to improve efficiency, stability and usability. The demo system is able to detect moving objects in dynamic scene efficiently during test. Thesystem is also extendable, it can be used as an independent moving object detectingand tracking system, while serve as a basic platform for researching on other fields ofcomputer vision is possible.