Improving Map Re-localization with Deep ‘Movable’ Objects Segmentation on 3D LiDAR Point Clouds

Abstract

Localization and Mapping is an essential component to enable Autonomous Vehicles navigation, and requires an accuracy exceeding that of commercial GPS-based systems. Current odometry and mapping algorithms are able to provide this accurate information. However, the lack of robustness of these algorithms against dynamic obstacles and environmental changes, even for short time periods, forces the generation of new maps on every session without taking advantage of previously obtained ones. In this paper we propose the use of a deep learning architecture to segment movable objects from 3D LiDAR point clouds in order to obtain longer-lasting 3D maps. This will in turn allow for better, faster and more accurate re-localization and trajectoy estimation on subsequent days. We show the effectiveness of our approach in a very dynamic and cluttered scenario, a supermarket parking lot. For that, we record several sequences on different days and compare localization errors with and without our movable objects segmentation method. Results show that we are able to accurately re-locate over a filtered map, consistently reducing trajectory errors between an average of 35.1% with respect to a non-filtered map version and of 47.9% with respect to a standalone map created on the current session.

A Survey on Localization for Autonomous Vehicles

“…One challenge for map generation is the presence of dynamic objects and temporary obstacles, which require continuous updating of the map. [100] used a deep learning architecture to construct a durable map from 3D LiDAR data by filtering removable objects based on the convolutional dual-view architecture, which helps the ego vehicle for LiDAR-based re-localization and trajectory estimation. Another example of object detection and removable object substitution by the CNN is shown in [101].…”

Deep Learning Sensor Fusion for Autonomous Vehicle Perception and Localization: A Review

“…In an attempt to improve the accuracy of localization, in [167] Vaquero et al suggested improving the quality of the prebuilt map first. They proposed segmentation of the dynamic moving objects in the map, such as other vehicles and pedestrians, in order to obtain a map that is valid for use for a longer period.…”

Dynamic to Static Lidar Scan Reconstruction Using Adversarially Trained Auto Encoder

“…We use Cartographer (Hess et al 2016) Baselines: We evaluate our model with two baselines: (1) A unprocessed dynamic frame (Pure-Dynamic) and ( 2) With an efficient LiDAR pre-processing step for handling SLAM in dynamic environment which removes the dynamic points from the LiDAR scan (Detect & Delete) (Ruchti and Burgard 2018;Vaquero et al 2019).…”