Visual tracking using adaptive color histogram model

Color provides a useful cue for image analysis and object recognition in robotics and automation applications. In most color based target recognition systems, the color models are invariant in a-priori and are never adjusted while the illumination condition changed. In this case, the system recognition is prone to error due to the change of surrounding illumination. To solve this problem, we have proposed a modeling method with high-tolerance based on probability distribution. The model is adaptive based on the change of illumination condition through on-line adjustment of the model parameters. We called this vision system an adaptive color vision system (ACVS). In this paper, we describe this ACVS system in detail and demonstrate it through the application of the histogram backprojection algorithm. We have conducted experiments which demonstrate the color object tracking by ACVS in a natural environment is an adaptive robustness and flexible system.