A multi-target detection framework for multirotor UAV

Multi-target detection for multirotor reconnaissance UAV is the focus of research recently. This paper designs and implements a multi-target detection framework for multirotor reconnaissance UAV. Image preprocessing algorithm consists of distortion correction, image enhancement, and image stitching are focused to solve the buried features and limited resolution of images, which is mainly caused by confusion and camouflage of targets with the background as well as high viewing distance. YOLOv3 deep convolutional neural network is used for target detection. Relative distance between targets and a ground sample of known size is roughly estimated. The framework outputs a panoramic image covering the mission field with detection results. Simulation experiment shows that image preprocessing algorithms improve the performance by increasing the accuracy of target detection. Flight experiment verifies the framework achieves good results in the target detection mission.