Automated behavior and statistical analyses from traffic images based on precise vehicle tracking algorithm

Abstract

Supporting efficient and safe drive is one of the most important issues on ITS. Therefore, it is necessary for traffic planning or traffic control to acquire traffic statistics and analyses behaviors of vehicles in detail. Previously, since such acquisitions and analyses have been performed manually, huge people and periods had been necessary. For the purpose of such statistics acquisitions and behavior analyses, traffic image analyses would be effective because that images contains more rich information that spot sensors. However, for many years, vehicle tracking in traffic images has suffered from the problems of occlusions. In order to resolve such a problem, we have been proposing the spatio-temporal Markov random field model (S-T MRF) for segmentation of spatio-temporal images. By using such a precise tracking algorithm, we automatically acquired traffic statistics from traffic images for twelve months. Thus, the system could so stably acquire directional vehicle counts, velocity of each vehicle, Quantity-Velocity plots, and so on. Consequently, such statistics would be very important for traffic engineers.