Accurate and Computationally-inexpensive Recovery of Ego-Motion using Optical Flow and Range Flow with Extended Temporal Support

Abstract

The ego-motion of a moving RGB-Z sensor is computed using combined range flow and optical flow pixel constraints. The emphasis of the work is on computationally cheap yet accurate estimation of inter-frame translation and rotation parameters using a linear small rotation model. To ensure accurate inter-frame motion estimates, an iterative form of the estimator is developed which repeatedly warps the target frame and measures its misalignment with the current frame. As these motion estimates are integrated temporally, to minimise drift in pose over time, additional temporal constraint is provided through the use of anchor frames. The algorithm is evaluated on the recently published TUB RGB-D Benchmark which also includes a set of standard metrics. Results presented suggest that performance is commensurate with alternative methodologies such as SLAM but at a fraction of the computational cost.

An evaluation of real-time RGB-D visual odometry algorithms on mobile devices

“…More recently, Fang and Zhang [13] compared different open-source VO implementations: Libviso2 [15], Fovis [20], DVO [23], FastICP, Rangeflow [22], 3D-NDT [1], CCNY [11] and DEMO [63]. The evaluation is performed on two scenes of the TUM dataset and on a challenging dataset created by the authors with illumination changes, fast motion and long corridors.…”

Experimental Evaluation of RGB-D Visual Odometry Methods

“…However, in recent years researchers have begun to propose fast ICP-based registration methods. The fourth kind is based on depth image and range flow theory [25], [7]. The key idea of these methods is similar to dense visual odometry using RGB images.…”

“…However, in recent years researchers have begun to propose fast ICP-based registration methods. The fourth kind is based on depth image and range flow theory [25], [7].…”

“…Understanding the accuracy changes/reductions of these methods with respect to different challenges is important, and helps us design a more robust motion estimation method for steering an autonomous vehicle. This paper experimentally compares several real-time RGB-D visual odometry methods including Libviso2 [3], Fovis [4], DVO [5], FastICP [6], Rangeflow [7], 3D-NDT [8], CCNY [9], DEMO [10]. It should be noted that the study of this paper only focuses on visual odometry methods, and therefore various RGB-D V-SLAM algorithms [11], [12] are not considered.…”

Real-time onboard 6DoF localization of an indoor MAV in degraded visual environments using a RGB-D camera

“…Since we want to get a very fast odometry estimation with low CPU usage, we do not always incorporate the visual information into the depth image based odometry estimation method. Though joint optimization methods [20], [21] may get more accurate estimation, it is much more expensive to compute. In our system, we are not too concern with the accuracy of odometry estimation method since our localization algorithm can correct the drift of visual odometry.…”