Distinguishing Unmanned Aerial Vehicles (UAVs) from birds and other flying objects in 3D LiDAR point clouds
Abstract
The goal of the work in this research project is to investigate methods to recognize small-sized Unmanned Aerial Vehicles (UAVs), e.g. quadcopters, in 3D LiDAR (Light Detection and Ranging) point clouds measured by a rosette scanning line LiDAR. Besides distinguishing between background and foreground elements, a major problem is filtering out false-positive detections due to birds and other flying objects. Swarms of insects and swirling dust can also negatively affect the measurements. Once detected, tracking the potential targets can also become a difficult task. We propose to tackle these challenges by researching new machine learning algorithms for detection and tracking. For this purpose currently no suitable dataset is available, thus during the first phase of our work we will create a labeled dataset that will serve as an input for the subsequent tasks. Our research will include investigating exploratory particle update strategies for a novel tracking algorithm, e.g., space partitioning to enable multi-target tracking and advanced background modeling (foreground-background segmentation), probabilistic motion model-based pose estimation. Analyzing measurable features that can potentially contribute to efficiently distinguishing UAVs from other flying objects (e.g. birds) is also part of the scientific challenge. We also aim to determine the feasibility of detection and tracking in case of partial occlusion, when temporary loss of information on the potential target further complicates the task.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 22, 2024
- Source ID
- FA86552317071
Entities
People
- Andras Majdik
Organizations
- Air Force Office of Scientific Research
- United States Air Force