3D LiDAR Map Compression for Efficient Localization on Resource Constrained Vehicles

Abstract

Large scale 3D maps constructed via LiDAR sensor are widely used on intelligent vehicles for localization in outdoor scenes. However, loading, communication and processing of the original dense maps are time consuming for onboard computing platform, which calls for a more concise representation of maps to reduce the complexity but keep the performance of localization. In this paper, we propose a teacher-student learning paradigm to compress the 3D point cloud map. Specifically, we first find a subset of LiDAR points with high number of observations to preserve the localization performance, which is regarded as the teacher of map compression. An efficient optimization strategy is proposed to deal with the massive data in original map. With the supervision of compressed map, a student model is built by training a random forest model fed with geometric feature descriptors of each point. As a result, the student model is able to compress the map without referring to the expensive numerical optimization. Additionally, by incorporating the features, the innovative student model can be generalized to other new maps while no re-training is required. We conduct thorough experiments on multi-session dataset and KITTI dataset to demonstrate the effectiveness and efficiency of the proposed learning paradigm, and the comparison with other map compression methods. The final results show that the learned student model can achieve efficient map compression with comparable LiDAR based localization performance to the original map at the same time.

Radar-on-Lidar: metric radar localization on prior lidar maps

“…which is similar with the observation count for map points in [22]. Based on the distance statistics on the P i t , the importance weights can be obtained.…”

CoFi: Coarse-to-Fine ICP for LiDAR Localization in an Efficient Long-lasting Point Cloud Map

“…It is obvious that our method outperforms LOAM, one of the best LiDAR odometry methods, using as low as 3.36% LiDAR frames, and needs as low as 0.01% points for the feature point map and 0.30% points for the long-lasting map. In comparison, the existing point cloud compression methods [64], [40], [65], [66], [67], [68], [69], [70] require at least 2% points.…”

“…However, point cloud maps are usually large in size and part of the points may change over the time. Several algorithm has been proposed to compress the point cloud map by compression [39], [40] or parametric representation [41], [42], [43]. To correct the changes in the point cloud map, several methods have been proposed to update the map during driving [36], [44].…”

AVP-SLAM: Semantic Visual Mapping and Localization for Autonomous Vehicles in the Parking Lot

“…To achieve robust localization, vehicles are equipped with various sensors, such as GPS, camera, Lidar, IMU, wheel odometer, etc. A great number of localization methods appeared over the last decades, such as visual-based methods [1,2], visual-inertial-based methods [3]- [5], Lidar-based methods [6]- [8]. For commercial-level production, low-cost sensors, such as IMUs and cameras, are preferred.…”

“…Global localization Relative localization is also called as odometry, which initializes coordinate at the start position, and focuses on the relative pose between local frames. Popular relative localization includes visual odometry [1,2,9]- [11], visual-inertial odometry [3]- [5,12], Lidar odometry [6]- [8]. On the contrary, the global localization has a fixed coordinate.…”