Research On The Identified Technique Of Image Based On Multicolor Pseudo-random Code

Machine vision is the eyes of automatization ,and is widely used in many fields, such as national economy,science research and national defense. One of the most important targets is to recover the scene of 3D data and to reconstruct 3D scene .In 3D reconstruction with machine vision method, a well-known problem is the ambiguity between the surface to be recovered and its image. Generally, in order to tackle this problem, a practical method is to use active machine vision techniques with structured light illumination. The active vision system with encoded structured light illumination is most effective, since the real Euclidean reconstruction of 3D scene can be implemented only with single image captured by uncalibrated camera whose parameters are unknown and allowed to change to adapt the varying scene. Therefore, researching the encoded structured light pattern to resolve the correspondence of coordination and image-decoded problem is a very important theory and technique problem in 3D scene reconstruction.This paper mainly introduces image identification technology on the base of pseudo-random color code, with the theory of uncalibrated 3D Euclidean reconstruction in machine vision, image processing methods and mathematical knowledge. The paper depicts the principle of pseudo-random code and its color encoded projection system that solves the image-decoded problem in 3D reconstruction employing corner detection to pick up the coordination of feature points in pseudo-random color encoded image captured by uncalibrated camera. Obtain a window of pseudo-random array made up of feature points on scene surface using color analysis. Then, employ neural network to calibrate the location of the encoded feature points in the whole template based on the unique window property of pseudo-random array. In this way, it solves the difficult identification problem of the correspondence between images of 3D surface reconstruction in machine vision, because each feature point can be exclusively identificated. The accuracy of capturing the coordination of feature points in pseudo-random color encoded image connects with the following accuracy of uncalibrated 3D reconstruction. The result of image identification is the restriction and direct basis of the following processing in machine vision.