Kernel Bandwidth Adaptive Target Tracking Algorithm Based on Mean-Shift

• Main Authors: Xiaofeng ZHANG, Dixing LI, Guowei YANG
• Format: Article
• Language: English
• Published: IFSA Publishing, S.L. 2013-07-01
• Series: Sensors & Transducers
• Subjects: Color feature; Motion detection; Adaptive kernel bandwidth; Mean-Shift algorithm; Kalman filter.
• Online Access: http://www.sensorsportal.com/HTML/DIGEST/july_2013/Special%20Issue/P_SI_416.pdf

The kernel bandwidth of the classical Mean-Shift tracking algorithm is fixed, and it usually results in tracking failure when the target's size changes. A kernel bandwidth adaptive Mean-Shift tracking algorithm is presented with frame difference method to solve the question in this paper. According to the targets' size obtained from the inter-frame difference method, the bandwidth matrix of kernel function can be updated. Because the improved Mean-Shift algorithm is difficult to position target while it moves fast, so a Kalman filter is put forward as auxiliary tracker further in the paper. The results of experiment show that the target's tracking accuracy of the non-rigid motion based this algorithm is improved by 3.28 % and the fast motion can be adapted to the target via the introduction of the prediction mechanism. This algorithm improved the defects on the target tracking which uses single color feature or motion information, so it is a practical tracking algorithm.