Towards Robust UAV Tracking in GNSS-Denied Environments: A Multi-LiDAR Multi-UAV Dataset

*University of Turku, +ETH Zurich

Abstract

With the increasing prevalence of drones in various industries, the navigation and tracking of unmanned aerial vehicles (UAVs) in challenging environments, particularly GNSS-denied areas, have become crucial concerns. To address this need, we present a novel multi-LiDAR dataset specifically designed for UAV tracking. Our dataset includes data from a spinning LiDAR, two solid-state LiDARs with different Field of View (FoV) and scan patterns, and an RGB-D camera. This diverse sensor suite allows for research on new challenges in the field, including limited FoV adaptability and multi-modality data processing.

The dataset facilitates the evaluation of existing algorithms and the development of new ones, paving the way for advances in UAV tracking techniques. Notably, we provide data in both indoor and outdoor environments. We also consider variable UAV sizes, from micro-aerial vehicles to more standard commercial UAV platforms. The outdoor trajectories are selected with close proximity to buildings, targeting research in UAV detection in urban areas, e.g., within counter-UAV systems or docking for UAV logistics.

In addition to the dataset, we provide a baseline comparison with recent LiDAR-based UAV tracking algorithms, benchmarking the performance with different sensors, UAVs, and algorithms. Importantly, our dataset shows that current methods have shortcomings and are unable to track UAVs consistently across different scenarios.


System Overview

Hardware Photo

The main contributions of this work and the presented dataset are the following:

  1. A dataset with data from 3 different LiDAR sensors and an RGB-D camera in both indoor and outdoor environments. This is, to our knowledge, the first diverse dataset in terms of LiDAR sensors for UAV tracking. The dataset includes a spinning LiDARs with 64 (Ouster OS1-64) channels, two different solid-state LiDARs (Livox Mid-360 and Livox Avia) with different scanning patterns and FoVs, and an RGB-D camera (RealSense D435). Given the short range of the point cloud generated by the camera compared to the LiDARs, we only extracted RGB images from it. Low-resolution images with depth, near-infrared, and laser reflectivity data from the Ouster sensor complete the dataset.
  2. The dataset includes sequences with motion capture-based (MOCAP) ground truth in both indoor and outdoor environments. The indoor trajectories exhibit more intricate patterns than the outdoors, while the outdoor sequences were deliberately selected to simulate potential docking and infrastructure inspection scenarios~\cite{seo2018drone} by emphasizing their proximity to a building.
  3. Based on the presented dataset, we provide a baseline comparison with recent LiDAR-based UAV tracking algorithms, benchmarking the performance with different sensors, UAVs of different sizes (from micro-aerial-vehicles to more standard commercial platforms), and algorithms
Our data collecting platform consists of a 64-channels Ouster spinning LiDAR (OS1), two Livox solid state LiDAR sensors: Mid-360, featuring a nearly 360° FoV, and Avia, with an almost-circular FoV. The setup is completed with an Intel RealSense D435 RGB-D camera.

Hardware Measurements
Our data collecting platform, top view (left) and front view (right)

Dataset

Our dataset is organized into three distinct categories based on the environment and trajectory structure: structured indoor, unstructured indoor, and unstructured outdoor. Each category captures specific movement patterns and characteristics, as follows:

  1. Structured Indoor: This subset (HolybroStdn) comprises simple trajectories represented by predefined, systematic patterns, including a circle, a cube, a spiral, and an up and down movement. These structured trajectories are intentionally included to provide standardized, reproducible, and easily interpretable movement patterns. By employing these basic trajectories, ablation studies can be performed on isolated and specific aspects of different methods. This naturally allows for evaluating scenarios where different elements are decoupled. The structured indoor trajectories act as a reference point for understanding how well a method performs under well-defined and controlled conditions.
  2. Unstructured: In this subset, trajectories exhibit a more irregular nature, simulating movements that occur both indoors and outdoors without strict adherence to predefined patterns. These trajectories aim to capture the spontaneous and less structured nature of real-world flight scenarios, where a UAV's movements can vary significantly based on the environment and other influencing factors.

Data format

The data collection makes use of the rosbag format. For each type of data included in the dataset, the data format is as follows:

  1. Point cloud data from spinning LiDAR (OS1-64) recorded as sensor_msgs::PointCloud. Each point in the point cloud contains four values (x, y , z, I), representing local Cartesian coordinates (x,y,z), and the measured laser reflectance (I).
  2. Point cloud data from solid-state LiDARs, Avia, and Mid-360, employs Livox's custom data format named livox_ros_driver/CustomMsg.
  3. Images from RGB camera at 1920×1080 resolution. The message type is sensor_msgs::Image.
  4. Images from the high-resolution spinning LiDAR (OS1-64) consisting of fixed-resolution range images, near-infrared images, and signal images.
  5. Inertial data from both spinning and solid-state LiDARs, featuring three built-in 6-axis IMU sensors with a 3-axis gyroscope and a 3-axis accelerometer. The standard ROS message type is sensor_msgs::Imu.
  6. Ground truth data from the MOCAP system and included as geometry_msgs::PoseStamped.

Download

Sequence Description Total Size Duration Difficulty Rosbag
HolybroStnd01 Structured (Up/Down) 8.5 GB 31.6s Easy Rosbag
HolybroStnd02 Structured (Square) 24.4 GB 90s Easy Rosbag
HolybroStnd03 Structured (Circle) 20.6 GB 76s Easy Rosbag
HolybroStnd04 Structured (Spiral) 26.5 GB 98s Easy Rosbag
Holybro01 Unstructured, Indoor 18.5 GB 68s Easy Rosbag
Holybro02 Unstructured, Indoor 19.4 GB 72s Easy Rosbag
Holybro03 Unstructured, Indoor 21.8 GB 81s Easy Rosbag
Holybro04 Unstructured, Indoor 20.8 GB 77s Medium Rosbag
Holybro05 Unstructured, Indoor 25.0 GB 93s Medium Rosbag
HolybroOut01 Unstructured, Outdoor 10.1 GB 37.8s Medium Rosbag
HolybroOut02 Unstructured, Outdoor 10.9 GB 40.6s Medium Rosbag
Autel01 Unstructured, Indoor 11.1 GB 41.4s Easy Rosbag
Autel02 Unstructured, Indoor 16.3 GB 60s Easy Rosbag
Autel03 Unstructured, Indoor 13.0 GB 48.6s Easy Rosbag
Autel04 Unstructured, Indoor 12.7 GB 47.3s Medium Rosbag
Autel05 Unstructured, Indoor 13.5 GB 50.2s Hard Rosbag
AutelOut01 Unstructured, Outdoor 10.1 GB 37.6s Hard Rosbag
AutelOut02 Unstructured, Outdoor 11.1 GB 41.3s Hard Rosbag
Tello01 Unstructured, Indoor 13.4 GB 49.9s Medium Rosbag
Tello02 Unstructured, Indoor 15.5 GB 57.8s Medium Rosbag
Tello03 Unstructured, Indoor 15.5 GB 57.8s Hard Rosbag
Tello04 Unstructured, Indoor 18.7 GB 69s Hard Rosbag
Tello05 Unstructured, Indoor 14.5 GB 54.1s Hard Rosbag
TelloOut01 Unstructured, Outdoor 10.3 GB 38.4s Hard Rosbag
TelloOut02 Unstructured, Outdoor 7.1 GB 26.4s Hard Rosbag
Calibration Office, Indoor 1.3 GB 9.3s Rosbag
The dataset is available on the University of Turku Servers

Data Sequences

Standard Trajectories


Tello


Autel


Holybro


Quantitative Results

Citation

@inproceedings{catalano2023towards,
      title={Towards robust uav tracking in gnss-denied environments: a multi-lidar multi-uav dataset},
      author={Catalano, Iacopo and Yu, Xianjia and Queralta, Jorge Pe{\~n}a},
      booktitle={2023 IEEE International Conference on Robotics and Biomimetics (ROBIO)},
      pages={1--7},
      year={2023},
      organization={IEEE}
    }