Exploiting a scene calibration mechanism for depth estimation

Abstract

The problem of depth estimation from one or more image(s) is most frequently discussed in computer vision using binocular cues, motion parallax or monocular cues. In this paper, we exploited a scene calibration mechanism for estimating depth from a single image, with emphasis on motorways. The approach incorporates linear perspective depth cue to restore distance information of vehicle(s) from a given image. Based upon the assumption that linear perspective is available in ample amount in structured environments, proposed approach computes 1D projective transformation across ground plane which maps imaged distances to the corresponding real-world distances. Once the homography matrix for 1D projective transformation is available, it can be applied to any point to compute its straight line distance from the reference point. Experimental results show that the proposed approach is computationally efficient and delivers desirably accurate depth estimates; thus, it has been applied to identify over-speedings.