UAV Object Detection and Tracking Advances

The field of UAV object detection and tracking is moving towards more robust and efficient solutions. Recent developments focus on improving the accuracy and reliability of detection systems, particularly in challenging environments with densely packed objects, scale variations, and occlusion. Innovations in feature encoding design, open-set detection, and multi-modal fusion are driving progress in this area. Noteworthy papers include: RT-DETR++ for its enhanced encoder component and channel-gated attention-based upsampling/downsampling mechanism. Model-Agnostic Open-Set Air-to-Air Visual Object Detection for its novel open-set detection framework and robustness against corrupted flight data. WAVE-DETR for its multi-modal approach combining visible RGB and acoustic signals for robust real-life UAV object detection. ISTASTrack for its transformer-based ANN-SNN hybrid tracker and effective feature interaction between ANN and SNN branches. T-SiamTPN for its temporal-aware Siamese tracking framework and significant improvements over the baseline. UCorr for its innovative solution to wire detection and depth estimation using a monocular end-to-end model.

Sources

RT-DETR++ for UAV Object Detection

Model-Agnostic Open-Set Air-to-Air Visual Object Detection for Reliable UAV Perception

WAVE-DETR Multi-Modal Visible and Acoustic Real-Life Drone Detector

ISTASTrack: Bridging ANN and SNN via ISTA Adapter for RGB-Event Tracking

T-SiamTPN: Temporal Siamese Transformer Pyramid Networks for Robust and Efficient UAV Tracking

Time-step Mixup for Efficient Spiking Knowledge Transfer from Appearance to Event Domain

UCorr: Wire Detection and Depth Estimation for Autonomous Drones

Built with on top of