Robust Real-time Car Tracking Unifying Low-level and High-level Tracking In a Stochastic Framework

点击翻译按钮获取中文摘要

Automatic traffic surveillance systems based on visual tracking techniques has been desired for many years, since it realizes many functions of ITS (Intelligent Transportation System). These functions includes estimating the number of the cars, speed, types of cars, and detecting some exceptional events (accidents, traffic jams). This technique may lead to the new services relating to the dynamic traffic information, course instruction or automatic drive operation. However, the robustness of the visual tracking is not sufficient for the practical systems.We have already developed an HMM-based background model for background- object-shadow separation and verified its validity on the low-level tracking system. In this research project, we intend to integrate the low-level tracker using the HMM-based background model with the high-level tracker which are able to track the deformation of the target object, into a probabilistic framework (Bayes' framework). This framework realiaes the practical tracking with the robustness to the change of light condition and to the rapid motion of the target object. We have designed and proposed the detailed framework of this mechanism. And we also discuss some problems such as the initialization of the tracking process and the possibility of the multi-object tracking, which are essential at the practical system design phase.This research project has been done for two years (April 2001 to March 2003). The result has been made public through the magazines of the academic societies, and the proceedings of the international conferences and other academic workshops.