The Study Of Face Tracking Algorithm Based On Surveillance Video

Face tracking is a classical problem in computer vision.It is applied in many important fields,such as video tracking,human-computer interaction,expression recognition and so on.Although great progress has been made in face tracking in recent years,most tracking algorithms are easy to be affected by illumination,occlusion,scale change and so on,which lead to performance degradation.Therefore,it is still challenging to implement accurate face tracking.In this paper,based on the previous research results,the traditional kernel correlation filter(KCF)and face appearance matching are improved,and the problem of target face tracking in surveillance video is deeply studied,and the results are as follows:1.As conventional kernel correlation filters tracking algorithm has poor performance in handing scale-variant heavy occlusion,scale adaptive kernel correlation filter face tracking with multiple features fusion is proposed in this paper.Firstly,the algorithm employs color attributes and Histogram of Oriented Gradient(HOG)to improve the face feature model,face translation is located by using a multiple channel filter.Then,learning an independent scale filter to estimate the optimal scale.Finally,using the method of linear interpolation to update the filter coefficient and facial appearance model.The experimental results show that the proposed algorithm significantly improves the performance of the face tracking.Both qualitative and quantitative evaluations show that the proposed algorithm is strongly robust to scale variations,heavy occlusion,while running at a real-time tracking speed of 36.3 frame/s.Compared to the existing tracking approaches,the proposed algorithm obtains better tracking performance.2.To address the problem of appearance matching across different challenges while doing visual face tracking.In this paper,face tracking is proposed that utilizes multiple appearance models with its long-term and short-term appearance memory for efficient face tracking.Firstly,it represents a face in a L2-subspace with a relational graph.Then,it utilizes pairwise distances of matched key points between consecutive frames that tackles drastic scale appearance change of a face.Finally,a weighted score-level fusion strategy is proposed to obtain the face tracking output having the highest fusion scores,and update the model through occlusion-detection.The experimental results demonstrate robustness to deformation,rotation,scale,illumination and background clutter,the tracker showcases impressive performance by outperforming many state-of-the-art tracker.