Fast Cylinder and Plane Extraction from Depth Cameras for Visual Odometry

Abstract

This paper presents CAPE, a method to extract planes and cylinder segments from organized point clouds, which processes 640×480 depth images on a single CPU core at an average of 300 Hz, by operating on a grid of planar cells. While, compared to state-of-the-art plane extraction, the latency of CAPE is more consistent and 4-10 times faster, depending on the scene, we also demonstrate empirically that applying CAPE to visual odometry can improve trajectory estimation on scenes made of cylindrical surfaces (e.g. tunnels), whereas using a plane extraction approach that is not curve-aware deteriorates performance on these scenes.To use these geometric primitives in visual odometry, we propose extending a probabilistic RGB-D odometry framework based on points, lines and planes to cylinder primitives. Following this framework, CAPE runs on fused depth maps and the parameters of cylinders are modelled probabilistically to account for uncertainty and weight accordingly the pose optimization residuals.

Real-Time Plane Detection with Consistency from Point Cloud Sequences

“…Finally, a pixel-wise region growing is executed to refine the extracted plane. Based on [ 12 ], a new method is proposed by Proena et al [ 19 ] for distinguishing cylindrical surfaces from the detected planes. When compared with [ 12 ], the latency of [ 19 ] is more consistent and 4–10 times faster, depending on the scene.…”

“…Based on [ 12 ], a new method is proposed by Proena et al [ 19 ] for distinguishing cylindrical surfaces from the detected planes. When compared with [ 12 ], the latency of [ 19 ] is more consistent and 4–10 times faster, depending on the scene. In [ 12 , 19 ], the pre-process is the key to making the algorithm run in real-time, which divides the range map into grid specific resolution, and it keeps a grid-level operation in most subsequent steps.…”

“…When compared with [ 12 ], the latency of [ 19 ] is more consistent and 4–10 times faster, depending on the scene. In [ 12 , 19 ], the pre-process is the key to making the algorithm run in real-time, which divides the range map into grid specific resolution, and it keeps a grid-level operation in most subsequent steps. This pre-process undoubtedly reduces the operation time in practice.…”

“…However, consecutive plane detection in videos could also assist those algorithms that require the correspondence between frames, such as adjacent point cloud registration in SLAM [ 16 , 17 ]. On-line methods, like [ 18 , 19 ], can establish plane correspondence, but the precision of segmentation is not satisfactory. Further, the frame-by-frame strategy would cause the “flicking” problem, as shown in Figure 1 .…”

Automatic Detection and Modeling of Underground Pipes Using a Portable 3D LiDAR System

“…The results showed an average error ranging from 10% to 18% at close distance measurements. Proença and Gao [22] proposed a method for fast plane and cylinder extraction to improve visual odometry performance on scenes made of cylindrical surfaces. In order to improve the efficiency and cater for sensor noise, image cells were used instead of 3D points.…”