The Study Of 3D Head Region Tracking And Depth Estimation

.With the development of science and the improvement of people’s safety consciousness, intelligent monitoring has more wide broad development. Compared to other parts, the information of head and face has a higher discrimination. So the importance of head detection and tracking head gradually become more and more crucial to the human identification recognition and improve the efficiency and performance of intelligent monitoring. From description of head region, the current detection and tracking of head have been gradually transited from two-dimensional information to the fusion of two-dimension and three-dimension. So what have a more far-reaching significance to intelligent monitoring, virtual reality and 3D animation movies are the studies of three-dimensional head region acquisition, processing method, and estimation of three-dimensional information based on two-dimensional information. The paper consists of two parts, head detection and tracking on depth image and depth estimation based on feature points of single face image.This paper conducted detection and tracking of the head region on depth image. Firstly, Head region detection system obtains the information of all objects in visual field, take distance transform to increase the details of all objects. Then utilize the improved template matching algorithm to detect all’head-like’regions in distance image detection. In order to improve the real-time performance of template matching algorithm, down-sample template, image and increase the stride length, which inevitably lead to rate of miss-matching. So this paper adopts the mathematical statistical analysis and take advantage of the candidate areas’features to extract the center of the head, at last the region growing algorithm is used to extract the whole header area.In head region tracking algorithm frame differential method and morphology processing are used to get the head region. Then use the improved Mean-shift algorithm for tracking. Mean-shift algorithm in this paper with a little improvement is applied to the depth of the image, which only use one channel information. At the same time forgetting factor is introduced to update the template. When angle between the target object and visual field center beyond a certain threshold, using the micro control system to control stepping motor to adjust Kinect. If no more than the threshold, it continues to track.In order to solve the situation of unable to obtain depth image directly, this article has been studied depth estimation, which is the critical problem of 3d head areas estimation. Specifically estimate the depth value of face image with using the existing 3d database. Firstly, establish the database on the basis of the existing 3d data, then adopt face++ program to detect the features on face region, and Delaunay triangulation is carried out. Thirdly, find out all triangle sub-region whose vertices include a certain point, dig out all triangles which is similar to triangle sub-region in the database and calculate the similarity. At last get depth value of features by weighted sum of Z axis value, and the weight distribution depends on the similarity between triangles.