Research On The Object Tracking Algorithmbased On Ranking Support Vector Machine

In recent years, visual tracking in computer vision has been widely used in many applications. As an active research topic, visual tracking has been extensively studied. However, it is still a challenging problem to track a target in real world environment because there are many influencing factors such as illumination and shape change, occlusion and clutter background.Visual tracking can be formulated as a learning to rank problem. In this dissertation, two visual tracking algorithms are proposed based on ranking support vector machine. The main contribution of this dissertation are as follows.Firstly, a visual tracking algorithm is proposed based on ranking support vector machine(RSVM) with multi-feature fusion. Two RSVMs are learnt with different features respectively and then fused with the weights calculating by the error rate of classifier. Then they are combined to be a robust visual tracking algorithm.Secondly, tracking problem is formulated as an online semi-supervised learning problem and a visual tracking algorithm is proposed in the co-training framework. Two RSVM are built with different types of features accordingly. Moreover, they are dynamically fused into a co-training process.Finally, the characteristics of infrared sequence and the methods of noise reduction on infrared image are studied. Then, the two proposed algorithms are tested on infrared sequences.Extensive experiments on challenging public available sequences and infrared sequences have demonstrated that the proposed tracking algorithm outperforms the state-of-the-art algorithms in terms of accuracy, robustness.