Large-Scale LiDAR SLAM with Factor Graph Optimization on High-Level Geometric Features

Abstract

Although visual SLAM (simultaneous localization and mapping) methods obtain very accurate results using optimization of residual errors defined with respect to the matching features, the SLAM systems based on 3-D laser (LiDAR) data commonly employ variants of the iterative closest points algorithm and raw point clouds as the map representation. However, it is possible to extract from point clouds features that are more spatially extended and more meaningful than points: line segments and/or planar patches. In particular, such features provide a natural way to represent human-made environments, such as urban and mixed indoor/outdoor scenes. In this paper, we perform an analysis of the advantages of a LiDAR-based SLAM that employs high-level geometric features in large-scale urban environments. We present a new approach to the LiDAR SLAM that uses planar patches and line segments for map representation and employs factor graph optimization typical to state-of-the-art visual SLAM for the final map and trajectory optimization. The new map structure and matching of features make it possible to implement in our system an efficient loop closure method, which exploits learned descriptors for place recognition and factor graph for optimization. With these improvements, the overall software structure is based on the proven LOAM concept to ensure real-time operation. A series of experiments were performed to compare the proposed solution to the open-source LOAM, considering different approaches to loop closure computation. The results are compared using standard metrics of trajectory accuracy, focusing on the final quality of the estimated trajectory and the consistency of the environment map. With some well-discussed reservations, our results demonstrate the gains due to using the high-level features in the full-optimization approach in the large-scale LiDAR SLAM.

Real-Time Detection of Non-Stationary Objects Using Intensity Data in Automotive LiDAR SLAM

“…Hence, in this paper, we focus on a relatively lightweight and computation efficient learnable filter for removing and/or weighting the range measurements depending on their assumed suitability for localization. The whole SLAM architecture we experiment with is built around a model-based approach [ 5 ], and we demonstrate that the accuracy of the estimated trajectories increases if we use our filter to eliminate the non-stationary objects.…”

“…The target application of our LiDAR data segmentation method is a selection of the points that are useful for SLAM, particularly in urban environments. The SLAM solutions we use in the presented experiments are the LOAM algorithm [ 8 ], considered the state-of-the-art in model-based LiDAR odometry, and our PlaneLOAM algorithm, which uses high-level features that group the measured points [ 5 ]. With a map consisting of planar patches and line segments, our system improves the accuracy of data association and can optimize the whole map using the factor graph SLAM formulation with the g 2 o library [ 43 ].…”

“…Moreover, all the accepted LiDAR points are used with the same weights, while Deschaud [ 44 ] demonstrated that keeping too many points that do not provide useful constraints (e.g., because they belong to movable objects or have large measurement errors) can degrade the accuracy of the computed transformation in LiDAR-based SLAM. Hence, in [ 5 ] we proposed robust data association methods for the creation and updating of the planar segment and line segment features, which allow PlaneLOAM to collect reliable LiDAR readouts in the form of high-level features. The feature-based map can be optimized, merging similar (multiplied) features and closing the detected loops, but it still may contain invalid features created upon LiDAR points originating from non-stationary objects.…”

“…Objects are considered as belonging to this category if they can move or are in motion, but also if their surface is not “rigid”, as a lawn area shaken by the wind, or tree canopies. The detection and segmentation of the LiDAR points belonging to such non-stationary objects are of particular importance in urban environments, where vehicles, pedestrians, and vegetation can be disregarded for localization, as there are enough man-made static surfaces [ 5 ]. A method that detects these objects in the context of automotive SLAM should process the incoming scans in real-time, using only the information that is yielded by the LiDAR itself.…”

“…While LiDAR intensity is used in SLAM [ 6 ] and place recognition to enhance the descriptiveness of LiDAR perception [ 7 ], this modality was so far not used for motion segmentation. Although our method uses already existing neural network architectures, we consider it an attempt to engineer a practical solution to a problem we have observed while developing PlaneLOAM [ 5 ], a feature-based LiDAR SLAM inspired by the LOAM (Lidar Odometry And Mapping) [ 8 ] processing pipeline.…”