Pole-based Vehicle Localization with Vector Maps: A Camera-LiDAR Comparative Study

TL;DR

This work tackles urban localization challenges under GNSS unreliability by leveraging pole-like road features referenced in cm-level vector HD maps. It compares camera-based pole detection against LiDAR-based detection and integrates detected poles into an EKF-based localization filter that fuses GNSS/DR with pole measurements and map features. The authors present a real-time, image-based pole detector trained with automatically annotated data and a geometry-based LiDAR pole detector, demonstrating that a multi-camera setup can achieve localization performance comparable to LiDAR on a challenging peri-urban sequence. Key contributions include a pole-based localization framework with map matching via Hungarian assignment, and an empirical evaluation showing that front cameras yield strong cross-track gains while side cameras boost along-track accuracy, with multi-camera fusion closely approaching LiDAR performance. The results support using multi-camera pole detection as a cost-effective alternative or complement to LiDAR for robust map-aided vehicle localization in feature-rich urban environments.

Abstract

For autonomous navigation, accurate localization with respect to a map is needed. In urban environments, infrastructure such as buildings or bridges cause major difficulties to Global Navigation Satellite Systems (GNSS) and, despite advances in inertial navigation, it is necessary to support them with other sources of exteroceptive information. In road environments, many common furniture such as traffic signs, traffic lights and street lights take the form of poles. By georeferencing these features in vector maps, they can be used within a localization filter that includes a detection pipeline and a data association method. Poles, having discriminative vertical structures, can be extracted from 3D geometric information using LiDAR sensors. Alternatively, deep neural networks can be employed to detect them from monocular cameras. The lack of depth information induces challenges in associating camera detections with map features. Yet, multi-camera integration provides a cost-efficient solution. This paper quantitatively evaluates the efficacy of these approaches in terms of localization. It introduces a real-time method for camera-based pole detection using a lightweight neural network trained on automatically annotated images. The proposed methods' efficiency is assessed on a challenging sequence with a vector map. The results highlight the high accuracy of the vision-based approach in open road conditions.

Figures (8)

• Figure 1: Image-based poles dectector training. A pre-trained semantic segmentation network is used to annotate an unlabeled custom dataset. The pixel-wise annotations are transformed into pointwise labels at the bases of the poles. A YOLOv7 is trained using bounding boxes centered at the pointwise labels. A small amount of manually labeled data are used for validation.
• Figure 2: Examples of segmented images and the obtained annotations (blue crosses).
• Figure 3: Examples of detections obtained from YOLOv7-based pole detectors on RGB and grayscale images. Each bounding box is displayed with its detection score and its center corresponding to a pole base is highlighted by a cross. For the grayscale side cameras, a final filtering is applied to remove detections on the vehicle roof.
• Figure 4: Examples of pole detections obtained from LiDAR. The bounding boxes of detected poles are visible in green. Two examples of missed detections are highlighted with black rectangles. Other examples are visible in point cloud distribution.
• Figure 5: Roof of the experimental Renault ZOE vehicle equipped showing the GNSS antenna and the Hesai Pandora sensor combining a LiDAR with several cameras.