Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Efficient Convex Hull-based Vehicle Pose Estimation Method for 3D LiDAR

Ningning Ding

TL;DR

This work tackles real-time 3D LiDAR vehicle pose estimation under sparse observations by converting 3D clusters to 2D convex hulls and evaluating pose with a minimum occlusion-area criterion. The method uses visible-edge selection and occlusion-area calculation to search the yaw orientation over a restricted range, achieving superior accuracy and speed on KITTI and a custom industrial dataset compared to RANSAC L-shape and CHH baselines. The approach yields mean absolute orientation errors around 1.7° on KITTI and ~1.3° on the authors’ dataset, with sub-millisecond per-object runtimes, supporting real-time deployment. These results suggest strong robustness to occlusion and sparse data, and the method is well-suited for obstacle avoidance and path planning, with future work to integrate tracking and outlier-resilient techniques.

Abstract

Vehicle pose estimation with LiDAR is essential in the perception technology of autonomous driving. However, due to incomplete observation measurements and sparsity of the LiDAR point cloud, it is challenging to achieve satisfactory pose extraction based on 3D LiDAR with the existing pose estimation methods. In addition, the demand for real-time performance further increases the difficulty of the pose estimation task. In this paper, we propose a novel vehicle pose estimation method based on the convex hull. The extracted 3D cluster is reduced to the convex hull, reducing the subsequent computation burden while preserving essential contour information. Subsequently, a novel criterion based on the minimum occlusion area is developed for the search-based algorithm, enabling accurate pose estimation. Additionally, this criterion renders the proposed algorithm particularly well-suited for obstacle avoidance. The proposed algorithm is validated on the KITTI dataset and a manually labeled dataset acquired at an industrial park. The results demonstrate that our proposed method can achieve better accuracy than the classical pose estimation method while maintaining real-time speed.

An Efficient Convex Hull-based Vehicle Pose Estimation Method for 3D LiDAR

TL;DR

This work tackles real-time 3D LiDAR vehicle pose estimation under sparse observations by converting 3D clusters to 2D convex hulls and evaluating pose with a minimum occlusion-area criterion. The method uses visible-edge selection and occlusion-area calculation to search the yaw orientation over a restricted range, achieving superior accuracy and speed on KITTI and a custom industrial dataset compared to RANSAC L-shape and CHH baselines. The approach yields mean absolute orientation errors around 1.7° on KITTI and ~1.3° on the authors’ dataset, with sub-millisecond per-object runtimes, supporting real-time deployment. These results suggest strong robustness to occlusion and sparse data, and the method is well-suited for obstacle avoidance and path planning, with future work to integrate tracking and outlier-resilient techniques.

Abstract

Vehicle pose estimation with LiDAR is essential in the perception technology of autonomous driving. However, due to incomplete observation measurements and sparsity of the LiDAR point cloud, it is challenging to achieve satisfactory pose extraction based on 3D LiDAR with the existing pose estimation methods. In addition, the demand for real-time performance further increases the difficulty of the pose estimation task. In this paper, we propose a novel vehicle pose estimation method based on the convex hull. The extracted 3D cluster is reduced to the convex hull, reducing the subsequent computation burden while preserving essential contour information. Subsequently, a novel criterion based on the minimum occlusion area is developed for the search-based algorithm, enabling accurate pose estimation. Additionally, this criterion renders the proposed algorithm particularly well-suited for obstacle avoidance. The proposed algorithm is validated on the KITTI dataset and a manually labeled dataset acquired at an industrial park. The results demonstrate that our proposed method can achieve better accuracy than the classical pose estimation method while maintaining real-time speed.
Paper Structure (9 sections, 1 equation, 8 figures, 2 tables, 3 algorithms)

This paper contains 9 sections, 1 equation, 8 figures, 2 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related work
  3. Methodology
  4. Visible edge selection
  5. Occlusion area calculation
  6. Experiments
  7. KITTI dataset
  8. Own dataset
  9. Conclusions

Figures (8)

  • Figure 1: Illustration of the proposed method. (a) A raw 3D point cloud cluster. (b) The extracted convex hull points. (c) Visible edge for a certain pose (The purple lines), the occlusion region's area will be calculated. (d) The final fitting result after optimization. (The red bounding box is obtained by our method. The green bounding box is the ground truth result.)
  • Figure 2: Schematic illustration of the visible edge selection and occlusion area calculation.
  • Figure 3: Sample results of the projection edge selection (The purple line denotes the projection edge). (a) Only one projection edge. (b) Two projection edges.
  • Figure 4: Distribution of absolute orientation error on KITTI dataset
  • Figure 5: Sample results of vehicle pose estimation on KITTI dataset. Green boxes from the ground truth, purple boxes from the RANSAC L-shape fitting method, blue boxes from the CHH method, and red boxes from the proposed algorithm.
  • ...and 3 more figures