Online Selection of Discriminative Pixel-Pair Feature for Tracking
Kenji Nishida, Takio Kurita, Masakatsu Higashikubo · 2010
A novel visual tracking algorithm is proposed in this pa-per. The algorithm plays an important role in a cooperative driving support system (DSSS) that is aimed at reducing traffic fatalities and injuries. The input to the algorithm is a gray-scale image for every video frame from a road-side camera, and the algorithm can be used to detect the existence of vehicles on the road and then track their tra-jectories. In this algorithm, discriminative pixel-pair fea-ture selection is adopted to discriminate between an im-age patch with an object in the correct position and image patches with objects in an incorrect position. The proposed algorithm showed stable and precise tracking performance when implemented in various illumination conditions and traffic conditions; the performance was especially good for the low-contrast vehicles running against a high-contrast background.