3D Reconstruction Based On Stereo Vision

3D RECONSTRUCTION IS A SIGNIFICANT FIELD OF RESEARCH IN COMPUTER VISION. ONE OF THE MOST USEFUL TECHNIQUES IS THROUGH STEREO VISION. GIVEN TWO IMAGES OF THE SAME SCENE, A PIXEL IN ONE IMAGE CORRESPONDS TO A PIXEL IN THE OTHER IF BOTH PIXELS ARE PROJECTIONS ALONG LINES OF SIGHT OF THE SAME PHYSICAL SCENE ELEMENT. DEPTH DATA OF MATCHES ARE USED TO RECONSTRUCT 3D MODELING. 3D RECONSTRUCTION IS PREVALENTLY USED TO DEVELOP THE COMPUTER VISION SOFTWARE FOR MOBILE ROBOT, MECHANICAL MANUFACTURE, CLINICAL IMAGE DIAGNOSIS AND SO ON.Based on Marr vision theroy, the five parts - system design, feature extraction, matching, camera calibration and visualization are discussed and analyzed. The 3D reconstraction is improved according to more recent achivement of stero vision. The main work is as follows:A STEREO VISION SYSTEM IS IMPLEMENTED, CONSISTING OF A PAIR OF CAMERAS(OR CCD), STANDARD OPTICAL PLATFORM, AN IMAGE ACQUISITION CARD AND A COMPUTER.A NEW EDGE-BASED MATCHING PROPAGATION ALGORITHM IS PROPOSED. IT STARTS FROM A SET OF KEY MATCHES TO NEIGHBORING PIXELS BASED ON A REGION GROWING PRINCIPLE AFTER INITIAL MATCHING. WE ALSO SIMULTANEOUSLY ENFORCE EDGE-BASED CONSTRAINT AND 2D DISPARITY GRADIENT LIMIT. THE PROPOSED METHOD CAN NOT ONLY IMPROVE THE MATCHING ACCURACY AND EFFICIENCY IN TEXTURE AREAS, BUT ALSO IMPROVE STABILITY IN HALF-OCCLUDED AND LOW-TEXTURED AREAS.AN ALGORITHM FOR FEATURE EXTRACTION IN 3D RECONSTRUCTION IS IMPROVED. COARSE EDGE INFORMATION IS DETECTED FIRST BY CANNY EDGE DETECTOR. KEY POINTS ARE THEN GENERATED WHICH UTILIZES THE GEOMETRICAL CHARACTERISTIC OF COUNTER. BY USING EDGE CONNECTION BASED ON LABELS DETERMINED BY KEY POINTS, THE COMPLETE EDGE IS EXTRACTED. FEATURE POINTS ARE ACCURATELY EXTRACTED BETWEEN KEY POINTS BY LINE SPLIT ALGORITHM.TWO MATCHING STRATEGIES, THE CROSS-CONSISTENCY CHECK AND THE BEST-FIRST STRATEGY, IN STEREO MATCHING ARE ADOPTED. AFTER THE CORRELATION STEP, CROSS-CONSISTENCY CHECK IS USED TO RETAIN A ONE-TO-ONE MATCHING BETWEEN TWO IMAGES. IN THE MEANTIME, MATCHES ARE SELECTED BY BEST-FIRST STRATEGY ONLY IF THEIR CORRELATION SCORES ARE RANKED ON THE TOP OF THE LIST. THE ROBUSTNESS AND ACCURACY OF MATCHING ALGORITHMS ARE IMPROVED.Experimental results have shown that 3D model reconstructed can reproduce vividly the surface of the object in different views through the ways of 3D graphic such as zooming, shrinking, shifting and rotation. The visualization is satisfaction.