Integrated Intensity, Orientation Code and Spatial Information for Robust Tracking

Object real-time tracking is an important topic in computer vision. Conventional single cue algorithms typically fail outside limited tracking conditions. Integration of multimodal visual cues with complementary failure modes allows tracking to continue despite losing individual cues. In this paper, we combine intensity, orientation codes and special information to form a new intensity-orientation codes-special (IOS) feature to represent the target. The intensity feature is not affected by the shape variance of object and has good stability. Orientation codes matching is robust for searching object in cluttered environments even in the cases of illumination fluctuations resulting from shadowing or highlighting, etc. The spatial locations of the pixels are used which allow us to take into account the spatial information which is lost in traditional histogram. Histograms of intensity, orientation codes and spatial information are employed for represent the target. Mean shift algorithm is a nonparametric density estimation method. The fast and optimal mode matching can be achieved by this method. In order to reduce the compute time, we use the mean shift procedure to reach the target localization. Experimental results show that the new method can successfully cope with clutter, partial occlusions, illumination change, and target variations such as scale and rotation. The computational complexity is very low. If the size of the target is 36 times 28 pixels, it only needs 12 ms to complete the method.