Analysis, Implementation and Application of Stereo Matching Algorithm for Image Depth Understanding

Stereo vision, resulting in the knowledge of depth information in a scene, is of significant importance in the field of machine vision, robotics and image analysis. Stereo matching is the process of computing a disparity map from a pair of stereo images. At the heart of every stereo matching algorithm is a solution to the correspondence problem – the problem of finding points in the right and left image that correspond to a single point in the real world. The paper presents a new method of feature based stereo matching algorithm. The state of the art algorithm for hybrid segmentation (HSAD) works fast and produce highly accurate depth map, but it is sensitive to different image conditions (i.e. illumination, rotation, blurring, scale, clutter, etc.). To overcome the problem, this research proposes an algorithm that utilizes OpenSURF algorithm and Hessian Matrix Laplace Function to enhance the image quality of HSAD matching. Speeded-Up Robust Feature (SURF) is a fast and performing scale and rotation-invariant interest point detector and descriptor. It uses a Hessian matrix for blob interest point extraction. Hessian matrix minimizes the filtering process and speed up the search for correspondence. The hybrid technique is integrated with the SAD stereo matching algorithm to determine the disparity estimate of each image pixel.