Research On The Optimization Of Head Pose Estimation Based On Interframe Information

Head pose estimation is a hotspot in the field of computer vision and human-computer interaction.The model-based head pose estimation method,according to the principle of pinhole camera model,the relationship between the two-dimensional image of the face and the three-dimensional model of the head is established,so that the head pose can be estimated by the change of the two-dimensional image.Under the condition that the location of feature points are accurate,the model-based method can obtain continuous estimates with high accuracy,and it is wildly studied.The head pose estimation method this paper studied is mainly divided into three steps:face detection,face alignment and parameter estimation.Face detection and face alignment are the front-end processing parts of the head pose estimation,which detect the face and provide the coordinate information of feature points for the parameter estimation.Parameter Estimation use iterative algorithm to get accurate head pose estimation.However,in this process,the slow speed of face detection,the drifting of feature points,the difficulty to get an accurate head three-dimensional model,etc.are still problems,how to quickly and accurately estimating head pose has been the direction of researchers' efforts.In this paper,the head pose estimation process in the existing problems,based on previous studies done the following:1.Based on the results of previous frames,the search strategy of face detection is improved,while ensuring the recognition rate of a substantial increase in detection speed.2.Optimize the selection of feature points,the previous frame pose angle is used as the precondition,and when the left and right deflection of the previous frame pose angle exceeds the threshold,part of the feature points of the current frame at the edge of the contour are considered as poor points and thus eliminated to improving accuracy.3.Optimize of the solving method to get head pose estimation,the matrix of the optimal parameters of the previous frame is directly used as the initial matrix of iterative refinement algorithm of the current frame,and the solving steps are simplified while ensuring the convergence precision.4.Use of back projection error to adjust the two-dimensional parameters of the common head model to improve the estimation accuracy.