Analysis of special transport behavior using computer vision analysis of video from traffic cameras

Abstract

Traffic analysis using computer vision methods becoming an important field in the traffic analysis research area. Despite this, common traffic models still rely on traffic planning methods, which treat all cars uniformly, meanwhile special vehicle can bypass traffic rules. Considering this, special vehicle can travel noticeably faster by avoiding traffic jams. This paper presents an analysis of special transport behavior case - movement by the opposite lane(MBOL). Our goal is to analyze under which conditions, such kind of specific traffic behavior happens and to present a regression model, which further can be used in special transport route planning systems or transport model simulations. The video from the traffic surveillance camera on the Nevsky Prospect (the central street of the city of Saint-Petersburg) have been used. To analyze traffic conditions (and detect MBOL-cases) we use well-established computer vision methods - Viola-Jones for vehicle detection and MedianFlow/KCF for vehicle tracking. Results show that MBOL happens under extreme main lane conditions (high traffic density and flow), with wide variety of parameters for the opposite lane.