IR Blob Target Tracking Based on Improved Mean Shift Method

The mean shift algorithm is a steepest ascent gradient method based on the feature of kernel density distribution. It has shown a good stability when tracking targets in color imagery. But the traditional mean algorithm doesn´t track well in IR imagery due to the lack of stable features such as color, texture, figure etc. To overcome this problem, a novel multi-kernel cascading scheme for mean shift algorithm is given so as to improve the feature expression of target in mean shift procedure, and the flexibility of algorithm is improved by combining with Kalman filter at the same time. The experiments performed on the Terravic Motion IR data set show the robustness and efficiency of the improved method.