Tracking Based on Better Feature Selecting

Using a specific single feature in the tracking framework named cam shift for specific environments can effectively track the target. We compute log likelihood ratios of class conditional sample densities from object and background to creating weighted image. The two-class variance ratio is used to evaluating how well they separate sample distributions of object and background pixels. In experiments, we take color histogram and edge histogram as candidate features. Along with changes in the environment, this feature evaluation mechanism will be embedded in a Cam shift tracking system that adaptively selects better discriminative feature for tracking. Examples are presented that this method is stable and robust.