Research On Depth Information Extraction Technology Based On Computer Vision

As a new field based on various interdisciplinary, the theory research and practical application of computer vision make great progress. Especially in recent years, the development of image processing technology and high performance computer coming into being provide favorable conditions for computer vision research and application. Therefore, the further research to computer vision has much theory meaning and practical value.Computer vision is to reconstruct the three-dimensional space information of an object from some of its two-dimensional images. Generally, one would like to reconstruct the object in three-dimensional space with only two different view points. The fundamental principle of computer vision with dual-camera has a strong resemblance to the visual perception procedure of mankind. We can get different images from two different view points and then calculate the position deviation of images pixels based on triangular measure. Finally we can reconstruct the three-dimensional coordinates of the object.The integrated computer vision system generally includes image acquisition, camera calibration, image pre-processing, features extraction, stereo matching and depth determination, composed of six parts, etc. We give prominence to stereo matching. A video capture system accessing to buffer directly is designed. Simultaneously, the calibration techniques of two cameras are studied. By using RAC (radial alignment constraint) of imaging process to decompose camera parameters and organizing the solving sequence of the parameters rationally, all parameters can be obtained through solving linear equations that avoid non-linear optimization. The pre-processing of image includes the noise filtering, the enhancement of image contrast, image edge detection, and so on. Based on the analysis and research of various pre-processing algorithms, the algorithms are implemented with VC++. During stereo matching, a two-stage stereo matching method using two-threshold-value judgment is modified based on edge feature. The procedure of the correspondence is fulfilled through the edge-detection, the similarity check and the consistence check. And based on double way matching, it acquires more best matching points. In this way, not only can we reduce the search for correspondences, but also ensure the validity of matching. Rebuild 3D image for disperse data and achieve depth information. As the result of the 3D reconstruction, the 3D model of the target is displayed in computer through OpenGL programming.This thesis has studied the crucial techniques of computer vision, including image acquisition, camera calibration, image pre-processing, feature point extraction, stereo matching, the calculation of 3D coordinates. The results of experiment with real images show that the methods used in this article are correct and feasible. This work provides a foundation for developing and constructing a while computer vision system.