Detection of Vehicle Flow in Video Surveillance

Abstract

Existing detection algorithms of vehicle flow in video detect moving objects by per pixel, so they may be corrupted by noises and the computational costs are high. In this paper, we propose a robust moving vehicle detection algorithm with background dictionary learning. An improved vehicle flow detection algorithm based on virtual regions and virtual lines is presented. To do this, we firstly divide an image into multiple image patches that have the same sizes. Each patch is an object or a background. Then, a background dictionary is learnt for each patch. The similarity between a patch and the background dictionary is measured, upon which a patch is distinguished as an object or a background. Additionally, the virtual detection line is used and combined into the virtual regions to detect the vehicles. Experimental results demonstrate the real-time and accuracy of the proposed detection algorithm.