An HMM-based segmentation method for traffic monitoring movies

J. Kato, T. Watanabe, S. Joga, J. Rittscher, A. Blake
2002 IEEE Transactions on Pattern Analysis and Machine Intelligence  
AbstractÐShadows of moving objects often obstruct robust visual tracking. We propose an HMM-based segmentation method which classifies in real time each pixel or region into three categories: shadows, foreground, and background objects. In the case of traffic monitoring movies, the effectiveness of the proposed method has been proven through experimental results. Index TermsÐCar tracking, hidden Markov model, image classification, image segmentation, wavelet coefficients. ae 1 INTRODUCTION THE
more » ... 1 INTRODUCTION THE main obstacle to robust visual tracking is that distracting features, such as clutter in the background regions, compete for the attention of the tracker and may succeed in pulling the tracker away from foreground (target) objects [3]. To make the tracker reliable, it is a common practice to discriminate the foreground pixels from the background pixels. Earlier researchers have attempted to increase the robustness of the tracker by image differentiation techniques [11], [2], [10], [13]. However, as to applications such as traffic monitoring systems, typical troublesome features are the shadows of vehicles, which are not wellhandled by traditional techniques. This paper introduces a new segmentation method [9] based on Hidden Markov Models (HMMs) to deal with problems of the shadows of vehicles. The method is mainly composed of two phases: the learning phase and the segmentation phase. In the learning phase, the tracking process learns the unknown HMM parameters with an EM-type (Expectation-Maximization) algorithm [5] over several seconds of a video sequence. In the segmentation phase, the process classifies each small region in a field image of a movie into three different categories: foreground (p), background (f), and shadow () over time.
doi:10.1109/tpami.2002.1033221 fatcat:gtrs6mmyrnb4rants4vwj2pn7m