A Study Of The3D Reconstruction Based On Omni Directional Stereo Vision And Sift Feature Points

This paper used single fold reflection Omni directional images and SIFT featurepoints to obtain three-dimensional information of the imaged object, for the mobile robotobstacle detection and navigation tasks. Catadioptric images are acquired by a Omnidirectional stereo vision system which includes a general perspective single-camera and avideo camera which consists of two hyperboloid mirror, system configuration makes twoimage point of the imaged object locate with one catadioptric Omni-directional image, themain point and image points of image plane are in a line. SIFT algorithm can find theextreme point in scale space; extract position, scale and rotation invariant.At first, a new similarity measure was designed for the feature point matching.Original catadioptric Omni directional image was expanded from cylinder and perspectiveprojection, improving the accuracy of matching by consistency check.Secondly, the stereo matching based on feature points and three-dimensionalcalculation can improve the accuracy and real-time of three-dimensional reconstruction,and due to SIFT feature points have scale and rotation invariance, the experimental resultsshow that, compared to the Canny edge feature points, based on SIFT feature pointsthree-dimensional reconstruction can obtain higher accuracy, the accrue of thethree-dimensional reconstruction can provide a basis for mobile robot obstacle judgment,has a good practical value. Therefore, in order to obtain the complete three-dimensionalmodel of the imaged object, stereo matching and3D calculations point is extended to theSIFT feature points.Finally, the paper designed three-dimensional reconstruction and depth estimation ofOmni directional image, showing image information in the form of three-dimensionalspace, can build the spatial location and the shape and size of various objects. Stereomatching and3D reconstruction of the feature points based on SIFT and catadioptricOmni-directional image show the reliability and practicality of the algorithm.