On-Road Moving Vehicle Detection by Spatio-Temporal Video Analysis of Static and Dynamic Backgrounds

Abstract

This paper is focused on addressing the challenges involved in building a single adaptive model of motion segmentation to detect moving vehicles from static and dynamic environments. The proposed algorithm identifies, extracts and removes the background components in two ways for both the cases. Initially, by dividing the frames into blocks and selecting the region of interest(s), i.e. moving traffic, where the vehicle presence is dominant in the scene. Second, a single method for background modelling and removal is used based on uniform spatio-temporal intensities. In a dynamic environment, the system makes use of Hough Transform to estimate and remove close connected components of the background in the spatial domain. In a static background, three-frame differencing technique proves beneficial for foreground detection. Additionally, efficiency of the system is improved by a constrained bounding box generation method. The results demonstrate the potential of the proposed system for publicly available aerial and acquired Bangalore Traffic datasets.