Denoising of Time-of-Flight depth data via iteratively reweighted least squares minimization

Abstract

Time-of-Flight depth data suffer from spatially varying noise, whose variance is inversely proportional to the squared amplitude of the received signal. On the other hand, preservation of genuine discontinuities of the scene is an important quality for a denoising method to have. This paper presents a noiseaware and discontinuity-preserving Time-of-Flight depth denoising method. To incorporate different constraints from the two philosophies, we recast depth denoising into an iteratively reweighted least squares problem, in which the cost function is iteratively updated and minimized in a manner of preserving the discontinuities and rejecting outliers while denoising the depth data. The experiments show that the proposed method delivers better results with lower error than existing methods, irrespective of the amount of noise and discontinuities.

Colored Point Cloud Registration by Depth Filtering

“…Our proposed method can be regarded as the unification of depth-error reduction [30,37] and point cloud registration [26]. Depth-error reduction algorithms refine measured depth values using the neighborhood within a depth image [30,37].…”

“…Our proposed method can be regarded as the unification of depth-error reduction [30,37] and point cloud registration [26]. Depth-error reduction algorithms refine measured depth values using the neighborhood within a depth image [30,37]. The Iterative K Closet Point (IKCP) algorithm [26] refines measured depth values using the K-closest points across point clouds.…”

Iterative K-Closest Point Algorithms for Colored Point Cloud Registration

“…Our depth refinement algorithm is similar to a depth noise reduction method [ 15 ] in that a depth-update equation is derived. The main difference is that our refinement is across point clouds, while the noise reduction in [ 15 ] is performed within a single depth image. In addition, the goal of our update equation is rather to align two point clouds more closely than to reduce the noise.…”

“…The first contribution of this paper is a novel probabilistic cost function based on the color-supported soft matching of points to their K -closest points. The depth values measured by RGB-D cameras suffer from errors [ 14 , 15 , 16 ], which hinder obtaining accurate one-to-one correspondences. The probabilistic one-to-many correspondences help to improve the pose accuracy in the presence of the measurement errors.…”