Object Detection and Tracking Using Spatiotemporal Wavelet Domain Markov Random Field in Video

In this paper, an object detection and tracking algorithm is proposed. At the object detection stage, the spatial features of object are extracted by wavelet transform, according to frame difference, the moving target is determined. In order to effectively utilize the temporal motion information, Markov random field prior probability model and the observation field model are established, taking advantage of Bayesian criterion, the location of object in successive frame is estimated, it contributes to the accurate space object detection. At the object tracking stage, using color and texture histogram of object, the Bhattacharyya coefficients are constructed, according to the color and texture Bhattacharyya distance, the joint likelihood tracker is established. In order to adapt the scene changes and enhance the tracker robustness, according to color and texture histogram, we construct Kalman filter with time-varying parameters. In the light of Kalman filter residuals statistical distribution, the candidate object model is updated by using hypothesis test. Experiments on international standard video sequences illustrate the effectiveness of the proposed schemes.