PSF-LO: Parameterized Semantic Features Based Lidar Odometry

Abstract

Lidar odometry (LO) is a key technology in numerous reliable and accurate localization and mapping systems of autonomous driving. The state-of-the-art LO methods generally leverage geometric information to perform point cloud registration. Furthermore, obtaining the point cloud semantic information describing the environment more abundantly will facilitate the registration. We present a novel semantic lidar odometry method based on self-designed parameterized semantic features (PSFs) to achieve low-drift ego-motion estimation for autonomous vehicle in real time. We first use a convolutional neural network-based algorithm to obtain point-wise semantics from the input laser point cloud, and then use semantic labels to separate road, building, traffic sign and pole-like point cloud and fit them separately to obtain corresponding PSFs. A fast PSF-based matching enables us to refine geometric features (GeFs) registration, thereby reducing the impact of blurred submap surface on the accuracy of GeFs matching. Besides, we design an efficient instance-level method to accurately recognize and remove the dynamic objects while retaining static ones in the semantic point cloud, which are beneficial to further improve the accuracy of LO. We evaluate our method, namely PSF-LO, on the public dataset KITTI Odometry Benchmark and rank #1 among semantic lidar methods with an average translational error of 0.82% in the test dataset.

Semantic geometric fusion multi-object tracking and lidar odometry in dynamic environment

“…Both refs. [12] and [13] using semantic segmentation networks realize semantic scene mapping. In addition, they detect dynamic objects through point-by-point checks from the object point cloud.…”

“…Refs. [12] and [13] use a semantic segmentation network [14] to obtain point-level semantic labels in the lidar scan and realize 3D map construction with scene semantics (roads, buildings). Undoubtedly, compared to schemes that check abnormal geometric constraints, SLAM systems which correctly introduce and use semantic information have more potential for application.…”

“…On the other hand, although the methods mentioned above can improve the robustness of the system in dynamic scenes, semantic SLAM systems such as refs. [12, 13] focus more on extracting and processing background semantic information. They cannot stably track and represent dynamic objects in the map coordinate, which is also necessary for many actual robot systems.…”

A Practical Survey on Visual Odometry for Autonomous Driving in Challenging Scenarios and Conditions

“…Other recent and relevant works that employ Deep Learning are PSF-LO [54], which uses parameterized semantic features to facilitate the registration task and employs a dynamic and static object classifier; and CAE-LO [55], that, like previous methods, uses unsupervised Deep Learning and utilizes compact 2D spherical ring projections. DMLO [56] is also an interesting work, making feature matching applicable to LiDAR odometry by decomposing the pose estimation in two parts: a matching network that makes correspondences between two scans and a rigid transformation estimation operation.…”

LiDAR Dynamic Target Detection Based on Multidimensional Features

“…These methods ensure a low data processing complexity, but typically use a single index as the benchmark to judge the motion state of the target, and different thresholds must be set for the corresponding indices in different scenes to obtain effective detection results. Methods based on learning segmentation generally use 3D−MiniNet, RangeNet++, and other networks to directly semantically segment LiDAR point clouds [ 13 , 14 ] or project point clouds into two-dimensional images for indirect semantic segmentation [ 15 , 16 , 17 ] and use the obtained semantic labels to detect dynamic targets. These methods are convenient to use and have a wide range of applications.…”