Hand tracking based on adaptive kernel bandwidth mean shift

摘要

In the process of human computer interaction, hand tracking is of great importance. A practical hand tracking method based on improved mean shift is proposed. Firstly, a rectangle tracking window containing the hand is determined manually in the initial frame. A target model based on the color histogram is established. For the subsequent frames, candidate models are also established. Next, the optimal center location of the target can be found by iterative operations with finite number of times. By modifying the radius of the kernel profile with a certain fraction, the size of the tracking window can be changed adaptively, and the proposed hand tracking method is no longer affected by scale changes of the target. Experimental results demonstrate that the proposed method can track the moving hand accurately. The proposed hand tracking method can be used in the fields of human computer interaction and augmented reality.