Depth detection of targets in a monocular image sequence

Abstract

This paper describes an approach for recovering structure of a moving target from a monocular image sequence. Within this paper, we assume the camera is stationary. We first use a motion detection algorithm to detect moving targets based on four heuristics derived from the properties of moving vehicles, maximum velocity, small velocity changes, coherent, and continuous motion. The second algorithm then estimates the distance of the moving targets using an over-constrained approach. We will show a proof-of-concept example from synthetic data. We have applied the approach to monocular image sequences captured by a moving camera to recover the 3D structure of stationary targets such as trees, telephone pole, etc. The experimental results on a monocular image sequence captured in an outdoor environment are also presented.