Research And Implementation Of Moving Object Detection And Road Detection In Traffic Video

Moving object’s detection is the fundamental technology for intelligent surveillance system, it has made many achievements after years of research. However, it still attracts the attention of many researchers due to the challenges brought about by the complexity of the real environment and the increase in associated demand.After the study of a variety of classical motion detection method, ViBe (Visual Background Extractor) is adopted as the means of implementation, Visual C and OpenCV as implementation platform, considering the accuracy and real-time performance. In addition, ViBe’s model initialization and background extraction are improved. Then, the problem is generalized, considering the scene mutation caused by PTZ camera’s change and light change, stationary of moving object, scene with pedestrians, etc, proposing corresponding appropriate solutions.Since moving object’s cast shadow moves together with the object and always differ much from the background, it could be recognized as moving target. It brings the subsequent processing a great deal of difficulties. So moving shadow detection becomes an important post-processing step to moving object’s detection.For the shadow detection, some classic feature-based method methods are studied and experimented in this paper, they are the HSV-based method, the LBP-based method, the edge-based method, the NCC-based method, and the latter three methods are improved. Then the improved LBP-based method and the HSV-based method is respectively combined with the HSV-based method to get better results.For the road detection, unlike conventional method that detection is performed in single frame, how to utilize the motion information to detect road is studied. Motion-area-based road detection and motion-trajectory-based road detection are proposed. For the case of single lane road, two-lane road, road with shadow, experiments are conducted, proving the robustness and practicality of methods. Finally, the complexity and applicability of the two methods are compared.