Traf: A Camera based System for Traffic Flow Analysis

Abstract

Abstract. As the traffic jam is becoming a serious issue of urban life, collecting traffic parameters for traffic distribution is important and critical. In this paper, we present a camera based system (Traf) to count automotive vehicles for traffic flow analysis use. The system employs the method of background subtraction by comparing differences between foreground and background. The subtraction result is utilized to identify the position and the shape of vehicles. The background model is initialed by averaging first 50 frames, and adapted to environment change by introducing an adapting parameter λ. Compared to state-of-art method, Traf is more robust, adaptive and economical.