Seed Point Acquisition Of 3D Facial Data Based On Binocular Stereo Vision

3D face recognition as an important branch of Biometric, has been a challenging task, and become a hot spot in the field of Biometric in recent years. 3D facial data as front steps of the entire 3D face recognition is very important. Seed point acquisition of 3D facial data is a critical step to 3D facial data acquisition, and also a crucial step to 3D face recognition.Based on the principle of binocular stereo vision and digital image processing technology, a method, which colored pattern is projected into the face to obtains seed point of 3D facial data, was raised. Firstly, project colored pattern onto human face. And then have snapshots from two different perspectives. Secondly, pre-process images, extract lines and feature points. Thirdly, match feature points and get feature point pairs. Finally, according to the principle of binocular stereo vision and camera parameters we calculate the coordinates of seed point. The advantages of this approach can be narrated as follows. Projected pattern contain a certain degree of texture, so facial texture goes complicated. The color increase contrast between feature points and background. The geometric characteristics of projected pattern provide geometric information for matching. This article elaborates the principle of binocular stereo vision, active optical projection technology, the corresponding point matching algorithm and 3D data reconstruction. The details are as follows.Binocular Stereo Vision is an important form of Machine vision, which is based on the principle of parallax and multiple images to obtain three-dimensional geometric information. The details of binocular stereo vision are discussed in Chapter 2. First of all, introduce the theory of binocular stereo vision. Secondly, the epipolar geometry and epipolar constraint of binocular stereo vision is elaborated. Thirdly, design the binocular stereo vision system architecture, including hardware and software components.The 3rd chapter is about active optical projection technology. First, introduce two prevalent types of active optical projection technology, their backgrounds and existing problems. Secondly, introduce the projected pattern. It is composed of the horizontal lines and the vertical line segments. The colors are composed of the green, blue, purple and back. Blue and purple block are marked points. Black background reduces the injury of projected pattern to human eyes. Feature points are intersections of line and line segment. See figure 1 for the colored pattern. Figure 1.A cut-out of the used pattern Then we talk about the image pre-processing, including color images sharp, color image contrast stretching, image smoothing and color images taken of the anti-gray color and so on. In the part of extraction lines, the LOG edge detection algorithm has been used. The advantage of LOG edge detection algorithm can get the fine edge of object, and consistency with the actual edge better. LOG edge detection algorithm also has anti-noise and potential anti-interference ability. The result of edge detection is the two boundary lines. Two boundary lines is not a requisite to this article, but the line itself. Therefore, edge detection algorithms have made great improvements to our goals. In the procedure, a threshold has been used. In order to enhance the stability of algorithms in this article, the threshold is auto-adapted. Because the lines thick different, we used thin algorithm in order to ensure the accuracy of feature points. In the section of feature point extraction, introduce the method of look for feature point. In this paper, the method of template is used to determine feature points. We get these templates by analysis of a large number of feature points. This method is simple and has high precision. Stereo match is a choke point of binocular stereo vision. Therefore the stereo match problem is related to the success or failure of this article. The details of stereo match are introduced in chapter 4. The outline of stereo match technology is introduced at first, including commonly used constraints and matching method, Such as epipolar constraint, unique constraint and the order of consistency constraint and the usage of image gray-scale of the similarity, feature points correlation method, dynamic programming method. Secondly, this article describes the matching program. The mark point is matched at first. So each line has its own number. Secondly, search for feature point and match along identical line of the left and right image. The order of consistency, unique and distance constraints is used in feature point's match. In this paper, the geometric features of the pattern used to achieve the feature point's match. Because without making use of the camera matrix, reducing the calculation error and improving the accuracy of matching.In the 3D reconstruction, this article is based on the principle of space a straight line intersect point for the space coordinates, two match points and optical center coordinates of the camera composition of the two straight lines, the intersection of these two straight lines that is what we are in need of. However, due to inevitable errors, two straight-line may be different, so the midpoint of common perpendicular of different line is what we need. In this paper, a method, which is based on binocular stereo vision projected colored pattern to obtain seed point of 3D facial data, is proposed. Although there are some problems, the principle is simple and with a lower requirement for level of equipment, it has a great prospective.