Object detection by clustering-based nonparametric kernel density estimation

Nonparametric model is one of the popular ones for background modeling for its ability to adapt to changes quickly in dynamic environment and enable very sensitive detection of moving objects. However, the method is too complex and computationally inefficient to real-time application. In this paper, a fast and efficient scheme for moving object detection by nonparametric background modeling was proposed, whereas the background was modeled combining clustering theory to eliminate the large amount of redundant information and noise pixels in background modeling stage. Moreover, adaptive background and temporal differencing was used to filter out unchanged background pixels, thus enabling an additional reduction in computational requirements and improving the detection results. Besides, RGB color components were substituted by chromaticity coordinates to suppress shadow effects. Experimental results demonstrated the effectiveness of the proposed algorithm.