Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unsupervised UAV 3D Trajectories Estimation with Sparse Point Clouds

Hanfang Liang, Yizhuo Yang, Jinming Hu, Jianfei Yang, Fen Liu, Shenghai Yuan

TL;DR

This work tackles the detection and 3D trajectory estimation of compact UAVs using sparse LiDAR data in an unsupervised framework. It introduces a global-local clustering approach combined with spatio-temporal density analysis and a voxel IoU-based scoring mechanism, followed by spline trajectory reconstruction to extract UAV paths without labeled data. Evaluated on the MMAUD v2/v3 dataset, the method demonstrates robust performance under day and night conditions and with sparse point clouds, highlighting its practicality for real-world, low-cost deployments. The study emphasizes open-source release and notes potential future directions, including integration of active countermeasures guided by the perception pipeline to mitigate UAV threats.

Abstract

Compact UAV systems, while advancing delivery and surveillance, pose significant security challenges due to their small size, which hinders detection by traditional methods. This paper presents a cost-effective, unsupervised UAV detection method using spatial-temporal sequence processing to fuse multiple LiDAR scans for accurate UAV tracking in real-world scenarios. Our approach segments point clouds into foreground and background, analyzes spatial-temporal data, and employs a scoring mechanism to enhance detection accuracy. Tested on a public dataset, our solution placed 4th in the CVPR 2024 UG2+ Challenge, demonstrating its practical effectiveness. We plan to open-source all designs, code, and sample data for the research community github.com/lianghanfang/UnLiDAR-UAV-Est.

Unsupervised UAV 3D Trajectories Estimation with Sparse Point Clouds

TL;DR

This work tackles the detection and 3D trajectory estimation of compact UAVs using sparse LiDAR data in an unsupervised framework. It introduces a global-local clustering approach combined with spatio-temporal density analysis and a voxel IoU-based scoring mechanism, followed by spline trajectory reconstruction to extract UAV paths without labeled data. Evaluated on the MMAUD v2/v3 dataset, the method demonstrates robust performance under day and night conditions and with sparse point clouds, highlighting its practicality for real-world, low-cost deployments. The study emphasizes open-source release and notes potential future directions, including integration of active countermeasures guided by the perception pipeline to mitigate UAV threats.

Abstract

Compact UAV systems, while advancing delivery and surveillance, pose significant security challenges due to their small size, which hinders detection by traditional methods. This paper presents a cost-effective, unsupervised UAV detection method using spatial-temporal sequence processing to fuse multiple LiDAR scans for accurate UAV tracking in real-world scenarios. Our approach segments point clouds into foreground and background, analyzes spatial-temporal data, and employs a scoring mechanism to enhance detection accuracy. Tested on a public dataset, our solution placed 4th in the CVPR 2024 UG2+ Challenge, demonstrating its practical effectiveness. We plan to open-source all designs, code, and sample data for the research community github.com/lianghanfang/UnLiDAR-UAV-Est.

Paper Structure

This paper contains 10 sections, 7 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Proposed Methods
  4. Global-Local Point Set clusterings
  5. Scoring Mechanism and Trajectory Prediction
  6. Experiment
  7. Dataset
  8. Evaluation Metrics
  9. Result and Discussion
  10. Conclusion And Future Works

Figures (3)

  • Figure 1: Illustration of detecting and tracking compact drones using a single low-cost sparse LiDAR to identify threats.
  • Figure 2: System Overview: Our algorithm uses DBSCAN to cluster point clouds, compares spatial-temporal changes, filters non-UAV data, and estimates UAV trajectories with spline fitting, measuring error with MSE.
  • Figure 3: This figure shows sampled points, ground truth, and our predicted trajectory, showing the accuracy of our solution.