A Pedestrian-Friendly Monocular Obstacle Avoidance Method

Abstract

A pedestrian interaction-friendly monocular obstacle avoidance method belongs to the field of unmanned aerial vehicle navigation. The invention solves the problem of poor obstacle avoidance performance of an indoor unmanned aerial vehicle equipped with a monocular camera. The method of the invention is that the UAV uses a monocular camera to collect pictures, the pictures are input into the parallel deep neural network structure of the end-to-end strategy, and the grid structure outputs the optimal heading angle as the flight instruction for the UAV to avoid obstacles; The parallel deep neural network structure of the end-to-end strategy is completed by the monocular camera combined with the single-line lidar. The training process is as follows: Step 1. Use the depth value collected by the single-line lidar to search for the best heading, and use the depth value collected by the monocular camera. Label the pictures to establish a data set; step 2, the data set is input into the Resnet18 network and the pre-trained YOLO v3 network respectively; step 3, use the data set of step 1 to train the parallel deep neural network in step 2 until convergence .

Description

technical field [0001] The invention relates to a Resnet18 deep neural network combined with a YOLOv3 deep neural network to form a parallel network structure to solve the monocular visual obstacle avoidance technology in a pedestrian scene, and belongs to the field of unmanned aerial vehicle navigation. [0002] Resnet (Residual Neural Network, residual neural network), YOLO (You Only Look Once: Unified, Real-Time Object Detection only look once: unified real-time object detection). Background technique [0003] With the development of the UAV industry, autonomous navigation of UAVs is the core of many UAV applications, such as in multi-UAV coordination, UAV mapping and UAV indoor tasks. However, due to the small indoor space and high personnel dynamics, the size of the drones used is limited, so the sensors that can be carried on small drones are also very limited (often only equipped with monocular cameras), so relying on limited Sensors that enable UAVs to autonomously ...

Images