Online object tracking based on L1-loss SVMs with motion constraints
Tao Zhuo, Peng Zhang, Yanning Zhang, Wei Huang · 2014
Orange technologies focus on individual behavior analysis, and the core of which is object tracking, especially arbitrary object tracking. One of the popular solution for arbitrary object tracking is tracking by detection. These approaches regard the tracking problem as a detection task, and use the online learning methods to adapt the classifier to various object appearance changes. However, due to lack of prior knowledge and unpredictable appearance changes, it is always hard to get accurate target location during the whole tracking process. In this paper, we incorporate a motion model into the tracking by detection framework. Besides object prediction, the motion model also guides the model updating process to guarantee the performance of the classifier. Experimentally, we show that our algorithm is able to outperform state of art trackers on benchmark data sets.