In this paper, training and testing of YOLO V3 in-depth learning object detection model were realized through acquisition and labelling of tarmac image data during flying process of unmanned aerial vehicle (UAV), and precise recognition of tarmac from UAV view was achieved at last, laying a good foundation for application of precise UAV landing. In this project, the most advanced visual in-depth learning object detection technology at present was applied, to recognize location of tarmac during landing process of UAV, thus enabling UAV to realize rough to precise positioning during the landing process, and realize civil UAV’s function of making precise recognition of tarmac and precise landing in an innovative way, which can also be operated and realized in lost-cost onboard small edge computing module.

Recognition of tarmac from UAV view based on Yolo V3

[1] FAN Qin-min. SSD Object Detection Based on Multi-layer Feature Fusion [D]. Sichuan: Southwest Jiaotong University, 2018.

[2] REN Shao-qing, HE Kai-ming, Ross Girshick, SUN Jian, Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks, arXiv: 1506. 01497

[3] RAN Xue. Algorithm Design and Realization of YOLO-based Object Detection [D], Chongqing: Chongqing University, 2020.

[4] LabelImg Open-source Software. [2021-11-09]. https://github.com/tzutalin/labellmg.

[5] PaddlePaddle YOLO V3. [2021-11-09]. https://www.paddlepaddle.org.cn/modelbasedetail/yolov3. [6] PaddlePaddle Paddle Lite. [2021-11-09]. paddlepaddle.org.cn/tutorials/projectdetai1/1978748.