3D target enhancement from single-shot polarization images via statistical outlier removal

Polarization images contain geometric information about detected target, which can be used for reconstructing three-dimensional structures with a specific process. Traditional polarization imaging systems are frequently hampered by sophisticated lens systems, which limits their portability and application versatility. With DoFP imaging system become increasingly compact, which incorporates a four-directional polarizer(0◦, 45°, 90° and 135°) on each pixel of the sensor. This allows the sensor to capture polarization information, in addition to traditional brightness and color data. In this paper, we combine polarization imaging system and point-cloud representation to present a high-contrast method for reconstructing three-dimensional structure of detected target. This system consists of a division-of-focal-plane polarization (DoFP) sensor carried a standard lens with 16mm and a white LED light source. Furthermore, point-cloud representation is better than mesh surface and voxel, especially in extract crucial information and denoising. We use a statistical algorithm named Statistical outlier removal (SOR) to remove noise. We tested two scenarios: reconstructing targets with simple structure and complex structure respectively.