Moving Object Detection Researchment And Hardware Implementation

Moving object detection is one of the most important processes in computer vision and moving image coding, has a vast perspective of application, it has high value in the theory of computer vision and has been a hotspot in the circle of international science. Moving object detection involves two phases of processes: the verification of an object in image sequences and locating it precisely. As the background is complex and changing in the scene, the precise detection of moving-objects is still a challenging problem.In this paper, some key techniques including moving object detection from image series which is acquired through a static monocular camera in specific traffic scenes are focused on. Some common methods in the field of object detection, such as Temporal Difference method, Background Subtraction method and Optical Flow method are discussed. The advantage and shortcoming of three algorithms are analyzed and compared firstly, according to different applications, the corresponding algorithm should be selected. Based on the studies of these algorithms, some improved algorithms are proposed and the design of the algorithm software module was realized are realized using Visual C or matlab.And according to characteristic of video images and the requirements of real-time for the system, the new sequence segmentation algorithm that extracts moving objects is introduced in detail in this paper. In this algorithm, the higher order statistics (HOS) hypothesis testing of inter-frame gray difference is used to automatically separate the motion region from background and morphologic method can be used to fill and smooth the gained binary mask. The algorithm software module programs are compiled using matlab. Experiment testifies that the proposed algorithm is of few parameters, robust to noise, with quick speed, compared with common methods mentioned above.In order to improve the efficiency of moving object segmentation furthermore, the HOS based algorithm is optimized as follows: Firstly, median filtering technology is used to optimize the image series in post process. Secondly, the threshold is found by gray level co-occurrence matrix considering the background texture change; Thirdly, in post process,a matrix-cluster approach and morphologic method are used to fill the binary mask. Experimental results demonstrate that the algorithm is best in result of segmentation and its real-time performance is excellent.Finally, the new system of moving object detection based on HOS algorithm is studied, this paper adopts the TMS320C6402 DSP produced by TI company as the primary control chip to design a hardware platform and the debugging result are gained.Seeing from the experiment result and data throughout the paper, the moving object can be exactly and timely detected by the system under the specified case and it can attain to the expected effect basically.