A grid-based histogram of the orientation gradient feature and visualization techinique were used in object tracking algorithm to solve the problem of tracking objects moving in surveillance video. The histogram of orientation gradient based on grid was designed as the local feature to describe the object. And it can apperceive the abnormalities of objects such as occlusion, disturbance and deformation etc. The visualization was used to real-time observe the changes of the characteistics through the changes of the local feature. The weights of the sub-model was adjusted adaptively to keep the stability of the tracking algorithm. The experimental results show that the proposed local feature can perceive the abnormity correctly, and can impove the robustness and stability of the tracking algorithm. The self-perception matching algorithm based on this local feature can tracking object accurately and stably in surveillance video.