VEHICLE TRACKING IN LOW-ANGLE AND FRONT-VIEW IMAGES BASED ON SPATIO-TEMPORAL MARKOV RANDOM FIELD MODEL
Shunsuke Kamijo, Katsushi Ikeuchi, Masao Sakauchi · 2001
One of the most important research in ITS is the development of systems that automatically analyze or monitor tra#c activities. For that purpose, it is necessary to achieve reliable vehicle tracking in tra#c images. However, occlusion e#ect among vehicles had impeded such reliable tracking for a long time. In order to solve this problem we have developed the dedicated tracking algorithm, referred to as Spatio-Temporal Markov Random Field in the year of 2000. This algorithm models a tracking problem by determining the state of each pixel in an image and its transit, and how such states transit along both the x-y image axes as well as the time axes. And it was proved that the algorithm has performed 95% success of tracking in middle-angle image. However, most of images actually captured by cameras on infrastructures are lowangle image, and many of them are front-view image. Since vehicles severely occlude each other in such images, segmentations of vehicle region through spatio-temporal images will be unsuccessful. In order to resolve such a problem, we have improved this ST-MRF model to re-optimize segmentation boundaries through accumulated spatio-temporal images. As a result, the improved algorithm were able to track vehicles at 91.2% success rate against such sever occlusions in low-angle and front view images at a highway junction. This successful result would lead precise analyses of severely complicated tra#cs.