Bilateral learning for color-based tracking

This paper addresses the issue of color model adaptation and color-based object tracking in a dynamic scene. Under different environmental conditions such as illumination changes, a static color model is inadequate for the purpose of color-based object detection and tracking. Color model adaptation is required and this has to be integrated into the tracking procedure within the spatial domain. To track a target in both the color and spatial domains, a Bilateral Learning (BL) approach is proposed in this paper. Formulated as an unsupervised learning problem, the adaptations of the color and spatial models are fitted into an EM framework by updating in the color and image spaces alternately. This results in the adaptation of the color model and the localization of the target along the image sequence. Experimental results show the effectiveness and efficacy of the proposed approach for color model adaptation and object tracking under illumination changes and environmental noise in real time.