Efficient Detection and Counting of Moving Vehicles with Region-Level Analysis of Video Frames

Abstract

The present article discusses the problem of detecting and counting of moving vehicle (MV) in a road traffic scenario, where background subtraction (BS) plays a vital role. BS in a video sequence is an open problem with many practical applications including camera surveillance system, human-computer interactions, etc. Among the various methods of BS, frame difference method is a simple and most adopted one. However, the performance of frame deference method depends on the proper selection of a set of frames. To meet this problem, we describe an efficient and fast processing approach for detecting and counting of MVs. A region/block-level analysis of frames is performed in this approach which requires less processing time and provides more accurate results compared to the conventional pixel-level analysis. We have used fuzzy flood fill mean shift based segmentation algorithm for this present study, which is robust under the illumination effects; such as shadows, shades, and highlights. In pixel-level analysis, segmentation operation is performed on the difference frame obtained from two test frames and detection of MV is made subsequently. However, detection with region-level analysis is made based on the geometric movement of regions between frames. Performance comparison between these two methods is made and superiority of the region-level based analysis is validated through examples. Also, we describe the estimation of the number of vehicles in a multiple MV traffic scenarios.