The Depth Estimate of Interesting Points from Monocular Vision

We present a novel method to estimate the depth and the spatial locations of points in the three-dimensional scene from monocular vision. First, we take photos when the camera is moving forward and obtain a sequence of photos of a scene. From this sequence of photos, we choose two photos. From these two photos, we can estimate the spatial position of points in the scene. Second, let a corner detector detect the corners on the latter image of the two ordered photos. Third, by using a novel method we can search the points on the former photo which match the corners on the latter image. Fourth, calculate the depths and the spatial locations of the points in the scene whose projections are the corners, by using a formula this paper will present. Compared with previous techniques estimating the depths or positions, our approach does not need any geometric clues, and the required conditions are that take photos while moving to the exactly forward direction. In addition, experiments are carried out to validate each proposed method and algorithm, and the experimental results demonstrate the efficiency of every proposed method and algorithm.