Research of UAV target detection and flight control based on deep learning

摘要

Deep learning has attracted widespread attention and has achieved good recognition results in image recognition and detection. Based on the Faster R-CNN algorithm, this paper studies the application of deep learning in drones and verifies its effect on target detection. In the most of existing schemes, the drone compresses the image and transmits it to the ground station for further detection, and then the resulting control command is transmitted back to the drone. This requires high bandwidth and high transmission delay, which limits the response speed of the system. In order to overcome the above shortcomings, the system will use the embedded system mounted on the UAV for image processing and pattern recognition, which saves transmission bandwidth and shortens response time. Through the Faster R-CNN target recognition algorithm, the UAV's target detection and flight control based on deep learning is finally realized. We performed HITL simulations of Pixhawk based on Gazebo in the ROS environment, and finally verified the feasibility of the algorithm.