Video Target Tracking Technology Research Based On Support Vector Machine

In recent years, visual object tracking is one of the core problems of computer vision, with wide-ranging application including human-computer interaction, augmented reality. As the video scene contains many complex information such as illumination, occlusion, background interference, coupled motion of the target itself, making the study of robust object tracking algorithm more difficult. In this paper, the visual tracking is implemented with structured output support vector machine(SVM) and the main research areas are as follows:(1)The classification algorithm based on SVM is implemented by deep study of different feature extraction methods, and gives the classification results. The redundant information in feature is removed by principle component analysis(PCA), and the experiment shows different application with different feature extraction methods.(2)Make an improvement on traditional structured output SVM, which overcome the target tracking drift caused by semi-occlusion or crosses. This paper proposes an fluctuant value of target match between frames to determine whether there is an exception, then to determine whether to update the support vector and combined with the Kalman filter to predict the target position in next frame for correcting tracking results.(3)As the temporarily left of target in video always bring error tracking in re-capture, this paper proposes improvements. when judging the abnormal of target and further judging its leaving, this paper proposes a timely warning and stop updating support vector, changing the search strategy until it can re-capture target.Experiments show that improved algorithm is more robust by comparing the original algorithm and offline algorithm, and has superiority when in response to more complex scenarios.