Improved stereo vision for extracting skeleton shape information in dense scatterer environments

Abstract

Three-dimensional (3D) reconstruction in dense scatterer environments is essential but challenging due to attenuation and scattering. Most specialized systems for this environment usually use a powerful, active illumination scanner, which can be slow and bulky. An attractive alternative is utilizing stereo vision, as it is typically a lighter and more cost-effective solution. However, stereo vision is an inherently ill-posed and challenging problem. It is even more difficult in the case of images of dense fog or dense steam scenes illuminated by active light sources. Images taken in such conditions suffer attenuation of object radiance and strong backscattering induced by the active light sources. Thus, the conventional well-developed stereo vision algorithms cannot be deployed directly to these images. In this study, we propose algorithms that improve stereo vision for extracting skeleton information in dense scattering media where the commonly used ranging sensors are unable to work. To solve this problem, we first derive the imaging model for images taken in a dense scattering medium illuminated by near-lighting. Based on the derived model, the non-uniform backscattering signal is efficiently removed from the images. The descattered images are then utilized as the input images of stereo vision. In the second method, the backscattering signal is eliminated from the measured signal when the stereo matching cost is pixel-wise calculated. The proposed algorithm focus on estimating a 3D cost volume by computing for each pixel in the left scatter-corrupted image a score for each possible disparity value. Then, the conventional well-developed methods can be deployed in cost aggregation and disparity estimation steps. We also demonstrate the effectiveness of the proposed method by extracting the skeleton shape information of the objects. From the 3D information of the object, the real robot manipulation task can be carried out.