On improving CAMSHIFT performance through colour constancy approach

Robustness is one of the key issues in single object tracking. It is hard to obtain a tracker that can perform well in various surroundings and environments. The complexity of the algorithm needs to be juggled with the intended processing speed. In this paper, we improved the robustness of single object tracker through a fusion with colour constancy approach. The aim is to transform the input image into a canonical form so that the lighting variation can be neglected. White path retinex is selected to analyze performance improvement for the selected CAMSHIFT tracker. A histogram based clipping pixel is used instead of a single pixel decision for obtaining the illumination source. The main innovation is the optimal fusion for transforming the image to a canonical form. Indeed, the original aim is to build a system that can perform well for uniform illumination only, yet it has exceeded our expectation as the results show improvement in non-uniform case too. For both test videos, the multiple object tracking precision index has improved by 8.43% and 32.35%. The algorithm is suitable to be implemented in many higher-level applications such as people monitoring, face recognition and computer human interface.