Moving Object Tracking in Occluded and Cluttered Backgrounds Using Adaptive Kalman Filtering

Abstract

This paper considers the problem of object tracking when a moving object undergoes partial or complete occlusion by the cluttered and noisy background. The presented algorithm is based on the Kalman filter and background checking combined with the mean shift algorithm. First, a rectangular region is defined surrounding the object of interest and the region is searched for a similar histogram distribution of that of the object of interest. Then, the model of the Kalman filter is constructed. Using the mean shift algorithm, the centroid of the object is predicted. The predicted values are fed into the Kalman filter. Interactively, the resulting parameter estimates of Kalman filtering are fed back to the mean shifting processor. The verification on the performance of the proposed method shows us that the proposed method can successfully track a moving object under complete or partial occlusion, even when the object has a similar color and texture with the background.