Research On Feature Points Location Of Multi-pose Face In Video

Facial feature points location has a wide range of applications in face recognition.It is the premise of face recognition research.In recent years,the research on facial feature points becomes a hot spot,but the accuracy of the location is the key to the study.In this paper,the ASM algorithm and L-K optical flow method were adopted to improve the accuracy of the locating of facial feature points and realize the tracking and positioning of the multi-pose facial features in video.The research covered the following aspects:Firstly,the common algorithms of the face feature points were analyzed,and the classification of them was summarized.The advantages and disadvantages of these algorithms were compared.Based on the research of the video,the tracking and localization of facial feature points in the video,the active shape model(ASM)algorithm was selected in the experimental study.Secondly,in order to solve the problem that the initial position of the ASM algorithm was not accurate,the Adaboost algorithm was used to detect the human face and found the approximate position of faces.Then ASM algorithm was used to locate the exact position.In addition,it acquired a lot of time and energy to mark the feature points of ASM model.A new method of correcting symmetric model was proposed,which can improve the time efficiency.Finally,in order to improve the positioning accuracy of the ASM algorithm,the method of image pyramid was introduced.Tirdly,in order to solve the limitation of a single ASM model in multi-pose faces locating,frontal face model,and the models of the two directions of the plane exterior rotation were trained.However,ASM method can not automatically determine the matching effect of the model when the face is located in the video,the L-K optical flow method is used to track the key points and the error between the feature points of the model is calculated.Fourthly,the method of affine transformation was used to improve the L-K optical flow method.The affine transformation parameters of the key feature points were calculated,and the tracking accuracy was improved.Then,the key feature points were tracked by the L-K optical flow method and the ASM algorithm,and the tracking accuracy and stability of the two algorithms were compared.Finally,the proposed algorithm was applied to the multi-pose faces video,and the experimental verification is carried out on the face of the trained ASM model.