Research On Vehicle Detection And Segmentation Method In Video

ABSTRACT:The technology of vehicle detection and segmentation is a very critical and fundamental part in the Intelligent Transportation Systems (ITS). It is an essential precondition to achieve the vehicle recognition, speed detection, grade identification and control of traffic flow. It involves several research domains, such as digital image processing, computer vision, pattern recognition and etc. MATLAB is used as the development platform and main research content of this dissertation is summarized as follows:(1) Background Reconstruction:an effective method, which is based on the hypothesis that background pixel value occurs most frequently, has been introduced firstly. In the method, successive three frames symmetrical difference algorithm was used to abstract the vehicle region in the rough state. In order to improve the precision of background beconstruction, three frames dissymmetrical difference algorithm that based on the rule to search the best frame interval is put forward in this paper and is used to take place of the successive three frames symmetrical difference algorithm in the original background beconstruction method. The experiment results show that the precision of background reconstruction has been greatly improved due to the three frames dissymmetrical difference algorithm.(2) Moving Vehicle Segmentation Precisely: Precise segmentation of the moving vehicle is a very critical and fundamental part in ITS. The original image segmentation method based on ESFCM (Edge-based Semi-Fuzzy C-means Clustering Method), which is named as original ESFCM in the paper, is introduced firstly. Then, three defects of the original ESFCM, which involves the great error in the initial subclass partition, the high computing complexity and the low antinoise performance, have been analyzed and summarized. In order to solve the three problems, an improved image segmentation method based on ESFCM, which is named as improved ESFCM in the paper, is put forword based on the supplement of the edge connection algorithm and redefinition of the dimensional distance. Later, the compare and analysis of computing complexity and antinoise performance between the original ESFCM and the improved one has been carried out. The experiment results shows that the image segmentation precision and the antinoise performance are improved and the computing complexity reduced after the the original ESFCM was substituted by the improved one. Lastly, the improved ESFCM with the great application value is used to realize the precise extraction of the moving vehicles in the vedio through the combined applications with the background difference algorithm.(3) Vehicle Activity Analysis:Based on the extraction of moving vehicle regions between two adjacent frames, the algorithm on vehicle activity analysis based on undirected bipartite graph is discussed in this paper. The overlapped area between every two vehicles regions in the two adjacent frames respectively is calculated out to be used as the parameter to judge that if the two regions are interrelated. Then, a vertex adjacency matrix that equivalent to the undirected bipartite graph will be achieved and simplified to represent the correspondence relations of every two vehicles regions in the two adjacent frames. According to the relations between the vertex adjacency matrix, moving region activity and vehicle activity, vehicle activity in the vedio has been realized based on the extraction of the characters of the vertex adjacency matrix.(4) Separation of the Occludded Vehicles:Occlusion in the monitoring video is a problem often encountered in the moving vehicles detection, tracking and identification. Based on the traditional method of four types feature points on contour, this paper proposes a new method of eight types feature points for the segmentation of vehicle occlusion. A rectangle region is assumed as the shape template of a vehicle. In this method, the first step is to extract the edge curve of the connected region, which has the phenomenon of vehicles occlusion, and then the feature points on contour were categorized into eight types. Subsequently, the feature points group, in which the feature points are the same kind and adjacent on contour, merged into an optimized one point according to the algorithm brought forward in the paper. At last, the vehicles were separated from each other by comparing every four adjacent feature points with the specific combination of the feature point types. The experiment results show that the new method of eight types feature points for the segmentation of vehicle occlusion is more robust, accurate, easily achieved and has good practical application merit.(5) Vehicle Speed Video Measurement:The traffic monitoring system based on the wireless Ethernet, which was designed by my project group, not only provides me with the factual video but also plays a part of experiment platform to test the algorithm designed in the paper. Vehicle speed measurement is carried out based on the camera calibration principle of Tsai's two stage method. The experiment results show that the vehicle speed measurement method in this paper is not only simple and practical but also highly robust and accurate. In a word, the speed measure method can fulfill the requirements in the vehicle speed video measurement system.