Multi-cue Fusion Based Depth Map Generation from 2D Video Frames

摘要

This paper proposes a novel depth map generation algorithm from 2D video sequences based on multi-cue fusion approach. Different from the traditional single depth cue generation method, the proposed multi-cue fusion method integrates several robust monocular depth perception cues and can eliminate the drawbacks of single cue. Besides, an improved aerial perspective generation algorithm based on dark channel prior is presented in this paper. Experimental results demonstrate that the proposed depth map generation method achieves encouraging performance, especially in the scenes with over bright regions, and the scenes with haze, smoke, or bubbles.