Robust Vehicle Tracking to Various Brightness Conditions and an Application to the Measurement of Traffic Parameters

Abstract

We propose a method for detecting and tracking vehicles in multilane traffic flow by using a camera installed on an overpass. The image intensities of stationary background scenes such as road surfaces and road markings are estimated through Kalman Filter based approach, and the feature points are detected only on vehicles by using Kanade-Lucus-Tomasi tracker. The Normalized Cuts based cluster analysis is utilized for segmenting the feature points into individual vehicles. Since a prior knowledge of vehicles as to shape, velocity and image intensity is unnecessary, the proposed method is robust to environmental changes in traffic scene, thereby increasing the possibility of an automatic measurement of traffic flow.