A comparison study on kernel based online learning for moving object classification

Most visual surveillance and video understanding systems require knowledge of categories of objects in the scene. One of the key challenges is to be able to classify any object in a real-time procedure in spite of changes in the scene over time and the varying appearance or shape of object. In this paper, we explore the applications of kernel based online learning methods in dealing with the above problems. We evaluate the performance of recently developed kernel based online algorithms combined with the state-of-the-art local shape feature descriptor. We perform the experimental evaluation on our dataset. The experimental results demonstrate that the online algorithms can be highly accurate to the problem of moving object classification.