Human Detection Based Yolo Backbones-Transformer in UAVs

Abstract

This study presents a new method for human detection in UAVs using Yolo backbones transformer. The proposed framework utilizes backbones YoloV8s, SC3T (Based Transformer), with RGB inputs to accurately perceive human detection. Experimental results demonstrate that the proposed method achieves an average accuracy of around 90.0% mAP@0.5 for human detection in the Human UAVs dataset, surpassing the performance of competitive baselines. The superior performance of our Deep Neural Network (DNN) can provide context awareness to UAVs. Furthermore, the proposed method can be easily adapted to detect UAVs in various applications. This work highlights the potential of the Yolo backbones transformer for enhancing human detection in UAVs, demonstrating its superiority over conventional methods. Overall, the proposed framework can pave the way for future research in UAV detection applications. Training code and self-collected Human detection dataset are released in https://github.com/Tyler-Do/Yolov8-Transformer.