An Adaptive Tracking Algorithm Based on Mean Shift
Zhenghong Deng,Tingting Li,Tingting Zhang
DOI: https://doi.org/10.4028/www.scientific.net/amr.538-541.2607
2012-01-01
Advanced Materials Research
Abstract:Object tracking is to search the most similar parts to targets in video sequences. Among the various tracking algorithms, mean shift tracking algorithm has become popular due to its simplicity, efficiency and good performance. This paper focused on mean shift tracking algorithm, which is a modeling mechanism based on statistical probability density function. In practice, when the background of the tracking and characteristics of the target are similar, pixels of background occupy a large proportion in the histogram. The traditional mean shift cannot adapt to the mutative scene. Meanwhile, if there is block or disappearance, the result is not exact. Three algorithms were given for above difficulties. A weighted template background was established, that can highlight the features of target and improve real-time. Then this paper presented a selective mechanism to update the target model. Every component is updated based on the contribution to the target model. Finally, the Kalman filter was combined with mean shift algorithm. We saw the prediction points of Kalman filter as the initial point, carried out the mean shift iteration and then updated Kalman filter using the ultimate location. Extensive experimental results illustrated excellent agreement with these methods.