An Algorithm of Moving Objects Localization Based on Neighboring Analysis

Accurate detection and location of moving objects is fundamental in surveillance application systems. Clustering methods are often applied to find the objects. However, this type of method is often confronted with some impediments, such as spoiled regions’ connectivity, a lot of outliers, great disparity between moving object regions, etc. In order to solve these problems, the chapter proposes a locating algorithm of moving objects based on neighboring analysis. Firstly, the neighboring relationship is analyzed between foreground pixels which are sampled from the background subtraction’s results, and a neighboring feature matrix is extracted. According to it, the initial clustering is obtained by linking the nearest samples. Furthermore, to minimize the criterion function, some clusters merged according to the relationship of the intra-class link function and the inter-class link function. Thus, all the samples converged into a conclusive number of clusters. The experiments indoors and outdoors demonstrate that it works effectively. The computational complexity is found to be proportional to the number of foreground moving objects.