Moving Target Detection Based On Genetic K-means Algorithm

In recent years, moving target detection and tracking has become a more and more hotter research topic. From the technical point of view, the topic belongs to the area of video process, while the research is related to many aspects, such as image processing, pattern recognition, computer vision; From another point of view of practical value, the study is the key technology in the construction of intelligent video surveillance system. No doubt that whether in military applications, or in the people’s daily life, intelligent monitoring has important application value. More importantly, it has a close relationship with the national economic development. Therefore, the study of the topic has important significance and broad application prospects.Moving target detection is not only the most critical part in the intelligent video system, but also the most influential part of the system effect. This paper has done much study and research on moving target detection algorithm, introduce three traditional methods:background subtraction, the frame difference and optical flow method. Exploring the research according to the principle of the theoretical basis from the following four perspectives:the comprehensiveness of moving target information, the interference immunity of the algorithm, the real-time effect of the algorithm and the effectiveness of dynamic scenes. And then a detailed summary of the advantages of each method was made. After all, a improvement was made based on the inter-frame difference method, which processing continuous three images. Through the experimental simulation, it comes out to have a nice effect.The most prominent of this paper is to achieve the topic from the clustering point of view based on Genetic K-means Algorithm. According the the theory of background subtraction, as long as the background model was established, moving target detection could be achieved. However, the commonly used Gaussian mixture model and modeling based on kernel estimation both have their own shortcomings. Considering the distribution of each pixel on the timeline, doing cluster process and then moving target and background were separated. In other words, the purpose of moving target detection was achieved. Through a detailed comparison of various clustering algorithms, the dominant position of Genetic K-means Algorithm (GKA) was established. According to the principle of the GKA algorithm, doing cluster analysis of the sample frame. After then, distinguish a pixel belongs to moving target or background according to the frequency it appeared. The moving target detection algorithm based on GKA has adaptability for dynamic background model; however, the algorithm could make updates and adjustments to the dynamic changes of the background, and a large sample of data acquisition and analysis or mass storage is needless.In addition, this paper also has done some research of the moving target tracking on the theoretical basis, while compared the performance of the Kalman filter and particle filter and achieved target tracking based on the Mean-Shift theory.