Research Of Face Detection And Tracking Algorithm

Face detection and face tracking technology is an important research direction in computervision. With the development of computer technology in recent years, the development of facedetection and face tracking technology is rapid, and it has been widely used in many fields.Especially it has some extensive and important applications in the fields of security surveillance,face search and human-computer interaction, which greatly promotes the modernization of thehabitat life. After investigation on domestic and foreign academic research on face detection andface tracking, several improved methods against the problems on face detection and tracking areproposed and experiments demonstrate the effectiveness of the algorithm. The main work of thispaper is as follows:(1) Improved skin segmentation algorithm based on Gaussian model is proposed. For colorimages with complex background and under different lighting division, an improved Gaussiandistribution which adapts to the lighting division in current image and the skin distribution of thecurrent image, reducing the missing detection rate and false detection rate of traditionalGaussian skin segmentation algorithm. Experiments conducted on faces database containingcolor images with complex background and under different lighting division demonstrate thevalidity of the algorithm.(2) Face detection algorithm based onAdaboost and improved Gaussian skin segmentationalgorithm is proposed. This algorithm combines theAdaboost algorithm with improvedGaussian skin segmentation algorithm. Firstly, candidate face regions in images are obtainedusingAdaboost algorithm. Then improved Gaussian skin segmentation algorithm is used toconduct skin segmentation on the candidate face regions. Non-face regions are weeded outaccording to the skin color pixel ratio in the candidate face regions. Experimental results showthat this method can effectively reduce the false detection rate of theAdaboost face detectionalgorithm.(3) Fragments-based face tracking using Compressive Sensing is proposed. As the Compressive Tracking algorithm is prone to tracking drift and tracking failure under occlusion,some improvements are made. Firstly, the candidate region is divided into several fragments.Then the classifier likelihood probability of the candidate region is obtained by weightedcombination of the probabilities of fragment classifier, which determines the target location.During the online classification update process, part of the fragment classifiers are updatedaccording to their weight in the previous frame to avoid the impact of occlusion in the previousframe on the overall classifier. Experiments on multiple video sequences show that theproposed algorithm can effectively improve the robustness of face tracking under occlusion.