Depth Recovery with Large-area Data Loss Guided by Polarization Cues for Time-of-Flight Imaging

Time-of-Flight imaging is one of the quintessential techniques in three-dimensional reconstruction. However, depth missing is a common problem in Time-of-Flight imaging, which can be classified into structure-based depth loss and large-area one according to various reasons. Large-area depth loss generally occurs due to close-range overexposure resulting in limited dynamic range in depth sensing. Compared to structure-based depth loss, the recovery of large-area depth missing is more challenging and has been rarely studied. In this paper, a large-area depth recovery framework guided by polarization cues is proposed stemmed from a solid physical basic concerning depth and polarization, to realize high dynamic range in applications. Inspired by RGB-D system and shape-from-polarization technique, a dual camera system is utilized including infrared Time-of-Flight camera and visible polarized camera. A physical model between depth map and polarization cues, specifically depth-gradient and degree-of-polarization, is investigated and established. Based on the physical basics, a corresponding polarization-guided depth recovery algorithm with statistical analysis and image processing approach is introduced. Experimental results towards different targets demonstrate the effectiveness of the proposed method qualitatively and quantitatively, accompanied with outcome analysis and detailed discussions about strengths and future works, which indicates a great potential for the applications of high-quality three-dimensional reconstruction and depth sensing.