Skip to Main Content
A new appearance model is presented for object tracking. The object model is built upon the color features and the distribution of object sub-regions into which the object is partitioned using k-means clustering. Thereby the updating of the object model is simplified as the updating of the object sub-regions. A similarity measure is introduced to evaluate similarity between the reference model and the candidate model. During object tracking, the parameters of the Kalman filter are adjusted and the object model is updated according to the update condition which indicates the degree to which the object is occluded. The results and analysis show that the proposed method has the ability to tracking the moving object real-time under real complex situations such as occlusion by other ones and change in object posture. The proposed method is an effective visual object tracking method.
Wavelet Analysis and Pattern Recognition, 2007. ICWAPR '07. International Conference on (Volume:2 )
Date of Conference: 2-4 Nov. 2007