Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Clutter Resilient Occlusion Avoidance for Tightly-Coupled Motion-Assisted Detection

Zhixuan Xie, Jianjun Chen, Guoliang Li, Shuai Wang, Kejiang Ye, Yonina C. Eldar, Chengzhong Xu

TL;DR

CROA addresses occlusion-driven failures in motion-assisted detection by tightly coupling detector viewpoint optimization with geometry-aware planning. It jointly minimizes occlusion probability and target-robot distance under polyhedron-based collision constraints, using a CCP-based nonconvex handling and a duality-based bilevel solver. The method is implemented for 3D lidar-based detection and validated in a CARLA-ROS simulator, where CROA reduces occlusion and detection errors while increasing the number of informative observations compared to planner- or detector-focused MAD schemes. The results indicate that accurate occlusion modeling and polyhedral obstacle representation yield substantial gains in cluttered, multi-lane driving scenarios, with practical impact for robust autonomous sensing and navigation.

Abstract

Occlusion is a key factor leading to detection failures. This paper proposes a motion-assisted detection (MAD) method that actively plans an executable path, for the robot to observe the target at a new viewpoint with potentially reduced occlusion. In contrast to existing MAD approaches that may fail in cluttered environments, the proposed framework is robust in such scenarios, therefore termed clutter resilient occlusion avoidance (CROA). The crux to CROA is to minimize the occlusion probability under polyhedron-based collision avoidance constraints via the convex-concave procedure and duality-based bilevel optimization. The system implementation supports lidar-based MAD with intertwined execution of learning-based detection and optimization-based planning. Experiments show that CROA outperforms various MAD schemes under a sparse convolutional neural network detector, in terms of point density, occlusion ratio, and detection error, in a multi-lane urban driving scenario.

Clutter Resilient Occlusion Avoidance for Tightly-Coupled Motion-Assisted Detection

TL;DR

CROA addresses occlusion-driven failures in motion-assisted detection by tightly coupling detector viewpoint optimization with geometry-aware planning. It jointly minimizes occlusion probability and target-robot distance under polyhedron-based collision constraints, using a CCP-based nonconvex handling and a duality-based bilevel solver. The method is implemented for 3D lidar-based detection and validated in a CARLA-ROS simulator, where CROA reduces occlusion and detection errors while increasing the number of informative observations compared to planner- or detector-focused MAD schemes. The results indicate that accurate occlusion modeling and polyhedral obstacle representation yield substantial gains in cluttered, multi-lane driving scenarios, with practical impact for robust autonomous sensing and navigation.

Abstract

Occlusion is a key factor leading to detection failures. This paper proposes a motion-assisted detection (MAD) method that actively plans an executable path, for the robot to observe the target at a new viewpoint with potentially reduced occlusion. In contrast to existing MAD approaches that may fail in cluttered environments, the proposed framework is robust in such scenarios, therefore termed clutter resilient occlusion avoidance (CROA). The crux to CROA is to minimize the occlusion probability under polyhedron-based collision avoidance constraints via the convex-concave procedure and duality-based bilevel optimization. The system implementation supports lidar-based MAD with intertwined execution of learning-based detection and optimization-based planning. Experiments show that CROA outperforms various MAD schemes under a sparse convolutional neural network detector, in terms of point density, occlusion ratio, and detection error, in a multi-lane urban driving scenario.

Paper Structure

This paper contains 8 sections, 12 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Proposed CROA Method
  4. Occlusion Probability
  5. CCP-Based Method
  6. Duality-Based Method
  7. Experiments
  8. Conclusion

Figures (5)

  • Figure 1: System model of robot MAD.
  • Figure 2: Occlusion probability computation of CROA.
  • Figure 3: Settings and motion profiles. In upper figures of (b)--(e), ground-truth and detected boxes are marked in red and blue, respectively. In middle figures of (b)--(e), robot trajectories, obstacles, and target are marked as red arrow lines, blue boxes, and green boxes, respectively.
  • Figure 4: Occlusion states of Fig. 2b--2e and OMPC failure.
  • Figure 5: Comparison of view qualities.