基于深度学习和机载LiDAR的无人机实时感知与探测
Real-Time Sense and Detect of Drones Using Deep Learning and Airborne LiDAR
AI总结：
本文提出首个基于机载LiDAR和3D深度学习的无人机集群实时感知与探测方案，通过垂直扩展扫描空间、加速稀疏卷积，并利用数字孪生增强数据，实现了高精度检测与间隔监控。
摘要：
无人机集群在超视距范围内的安全运行需要多重保障措施，以降低近距离飞行场景中无人机之间碰撞的风险。依赖预先规划的轨迹、持续的卫星和网络连接以及可靠的全球导航卫星系统（GNSS）定位的协同导航和飞行协调策略，在面对故障时较为脆弱。无人机嵌入式感知与探测为无人机之间的冲突消解和碰撞避免提供了一种全面的间隔模式。本文提出了首个基于机载LiDAR的无人机集群检测与定位解决方案，采用3D深度学习模型。该方案通过垂直扩展扫描空间，将现有的深度学习神经网络适配到空对空无人机场景中。本文提出并应用了一种新的稀疏卷积来加速骨干网络层，该层是神经网络中最耗时的部分。为了收集安全关键、近距离多无人机操作的训练数据，使用场景数字孪生以高保真合成数据增强真实数据集。训练后的模型在真实世界数据集上测试时，实现了超过80%的召回率和96%的精确率。通过结合基于检测的跟踪算法，该系统能够在具有挑战性的环境中可靠地监控多架无人机之间的间隔距离。
英文摘要：
The safe operation of drone swarms beyond visual line of sight requires multiple safeguards to mitigate the risk of collision between drones flying in close-proximity scenarios. Cooperative navigation and flight coordination strategies that rely on pre-planned trajectories, constant %{satellite and network connectivity and reliable Global Navigation Satellite System (GNSS) positioning are brittle to failure. Drone embedded sense and detect offers a comprehensive mode of separation between drones for deconfliction and collision avoidance. This paper presents the first airborne LiDAR based solution for drone-swarm detection and localization using 3D deep learning model. It adapts an existing deep learning neural network to the air-to-air drone scenario by expanding the scan space vertically. A new sparse convolution is proposed and applied to accelerate the backbone layer, which is the most time-consuming part of the neural network. To collect training data of safety critical, close-proximity multi-drone operations, a scenario Digital Twin is used to augment real datasets with high fidelity synthetic data. The trained model achieves over 80% recall and 96% precision when tested on real-world datasets. By incorporating a tracking-by-detection algorithm the system can reliably monitor the separation distance of multiple drones in challenging environments.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
