Robot Localization Technology Research Based On Monocular Vision

In recent years, computer vision has been widely used in various fields. The mainpurpose of the research of this paper is to introduce the computer vision to the automaticproduction line, in order to guide the manipulator to complete the work of work piecefetching. The major role of the computer vision in the production line is to obtain theinformation on the position and orientation of the work piece, namely allocation the targetwork piece. The introduction of computer vision greatly improves the automation degreeof the system.This paper uses a monocular vision system; relative to the binocular vision systemand multi-camera vision system this system has some advantages such as simple structure,cheap, easy calibration and so on. For the position of the camera is fixed in the practicalapplication or with a movement of the manipulator movement, this paper describes twolocalization algorithms. When the position of the camera is fixed, the localizationalgorithm is based on a single frame image; when the position of the camera is not fixed,the localization algorithm is based on the two frames. In the actual application can choosedifferent algorithm flexibly.The completed work of this paper mainly includes the following aspects:According to different application environment, this paper detailed introducedtwo monocular vision localization algorithms. Respectively, for the cases of the cameraposition is fixed or not. And gives the calculation processes of two kinds of algorithm,finally, has carried on a experiment to verify the feasibility of the two methods.Introduces the imaging model of camera, and the camera’s epipolar geometryprinciple. On the basis of the traditional method of calculating the basic matrix of epipolargeometry, this paper proposes a method named iterative minimization of optimal matching.This method improves the precision of the fundamental matrix greatly.This paper induces some knowledge about image processing: background subtraction,threshold, edge detection, corrosion, expansion, etc, and these knowledge were applied tothe image processing module of monocular vision system and achieved good results. Described several commonly used methods of point extraction. The characteristicpoints extracted by using Harris corner detection algorithm is uniform, reasonable andconform to the image structure; and the quantity of the characteristic points is large byusing SIFT algorithm but the match accuracy is high. This paper adopts the method ofcombining two algorithms to complete the matching of feature points.