Tracking Using CamShift Algorithm and Multiple Quantized Feature Spaces

The Continuously Adaptive Mean Shift Algorithm (CamShift) is an adaptation of the Mean Shift algorithm for object tracking that is intended as a step towards head and face tracking for a perceptual user interface. In this paper, we review the CamShift Algorithm and extend a default implementation to allow tracking in an arbitrary number and type of feature spaces. In order to compute the new probability that a pixel value belongs to the target model, we weight the multidimensional histogram with a simple monotonically decreasing kernel profile prior to histogram back-projection. We evaluate the effectiveness of this approach by comparing the results with a generic implementation of the Mean Shift algorithm in a quantized feature space of equivalent dimension. The aim if this paper is to examine the effectiveness of the CamShift algorithm as a general-purpose object tracking approach in the case where no assumptions have been made about the target to be tracked.