Research On Object Recognition And Localization Based On Stereo Vision

Unlike monocular vision,which is limited to 2D images of objects,the stereo vision system can obtain more abundant visual information;Thus,the stereo vision technology has a huge role in the avoidance of intelligent robots,object localization,object recognition,has been more and more attention and research.Especially in object recognition and localization technology,It can accurately identify object and locate three-dimensional coordinates of the object,If it is used in the identification and classification of industrial pipeline work piece and inspection and maintenance of petrochemical industry parts,which will have great application value to the society.This paper which put it as a background to research recognition positioning technology of stereo vision,to solve the current problems and optimize its shortcomings.The main work contents and innovative achievements are as follows:1.Camera Calibration: The calibration method of 11× 12 squares of the checkerboard calibration plate is used to calibrate the camera.The relevant parameters of the camera are obtained and the accuracy of the calibration results are verified,to facilitate follow-up study of three-dimensional reconstruction and positioning objects.2.Image pre-processing: combining experimental to study several image pre-processing methods.An image denoising method based on the combination of median filter and Gauss filter is proposed,which is to use a Gaussian filter for initial filtering,and then median filter to achieve the purpose of improving image quality,in order to facilitate the final image recognition.3.Stereo Matching:This paper focuses polar-line correction which is one of the constraint conditions of the stereo matching,the experimental results show the distortion of the image after the polar-line correction is obviously weakened,thus greatly speeding up the image stereo matching and 3D information recovery process.4.Location: The suitable value of the threshold value in stereo matching is studied,which is a reference value to measure the matching point,and it is applied to the object orientation in this research.Propose an indirect verification method validation positioning accuracy,that is for verifying positioning accuracy by distance between known corner points and corner points in calibration board.5.Object recognition: The SIFT feature descriptor is improved,respectively,from the scale,rotation and occlusion three aspects carries on the improved algorithm and the original algorithm experimental verification,by analyzing and comparing the experimental results,it is proved that the improved algorithm has the advantages of high matching accuracy and speed;Finally,a recognition and positioning experiment was carried out on a work piece.