The Researchand Implementation Of Moving Target Detection On Embedded System

Moving target detection is a image segmentation method that is used to detect the changearea and extract the interested moving target from the video sequence. It is a fundamental andkey technology in computer vision research, the result of which play a basic and key role insubsequent statistical analysis and objects’ behavior understanding and judgment of intelligentprocessing. In addition, with great enhance of embedded processor performance,it becomes atrend that image processing algorithm make a combination with the embedded platform that iswidely used in many area,such as intelligent transportation, intelligent household and intelligentcity. However, in view of the ever changing and complex environment,it is challenging toovercome the influence of illumination change and noise and then extract moving target moreaccurately. In addition, in view of the limited resources of embedded platform, puting forward areal-time and lower complexity algorithm is also the focus of this paper. Aiming at accuracy andreal-time two aspects,this paper make a more thorough research and discussion.To solve the problem of low efficiency in the process of codeword matching and lowaccuracy of detection under complex environment，this paper proposed a novel layered ordered-list codebook target detection algorithm which was based on regional characteristics analysis.Firstly，a newly proposed structure of codebook-codesection-codeword was used to hierarchizethe codebook. After optimizing the parameters of codeword, all of the codewords which weresorted according to the access frequency were organized into a fixed-length oredered list.Theexperimental data indicates that the strategy has greatly improved the performance of real time.Secondly，due to the shortcoming of overlooking relevance in pixel area，a moving-edge-markedmethod based on the local directional mode was designed to obtain the edge of moving objectsmore accurately and it also had better robustness in noise. In the last，the moving edge was fusedwith the segmented movement area of codebook. A stereoscopic model which reflects theessential attribute of the moving target in time and spacewas designed according to the algorithm.The results of experiment show that the proposed method has higher recognition rate and betterrobustness even in natural scene with illumination variation and random noise.The experimental results show that the algorithm can meet space and and time complexityrequirements of the embedded platform. It can process video data in real-time, and it can alsoeffectively overcome the influence of noise. And thus it accurately portray the essential attributeof moving targets, naturally achieve high detection accuracy.