A low-altitude small UAV obstacle perception method based on binocular vision

Abstract

The invention discloses a binocular vision-based obstacle perception method for a low-altitude small unmanned aerial vehicle, which belongs to the technical field of machine vision. First, the deep learning target detection and recognition method based on YOLOv2 is used to detect and recognize obstacles in the image collected by the left camera of the binocular camera, and obtain the pixel position, pixel size and obstacle type information of the obstacle in the image. Then, based on the above information, the KCF target tracking algorithm is used to track the target obstacles in real time, and at the same time, the three-dimensional reconstruction is performed on each frame of environmental images collected in real time by the left and right cameras of the binocular camera to obtain the spatial information of obstacles in the environment. Finally, combining all the above information, the obstacles in each frame of image are extracted, and the spatial position, physical size and obstacle type of all obstacles in the environment are obtained. The invention greatly reduces the complexity of data collection and production, can obtain rich obstacle information, and provides guarantee for the obstacle avoidance of the drone.

Description

technical field [0001] The invention belongs to the technical field of machine vision, in particular to a binocular vision-based obstacle perception method for low-altitude small UAVs. Background technique [0002] In recent years, drones have been used more and more in low-altitude fields, such as security monitoring, agricultural plant protection and power inspection, and other fields are playing an increasingly important role. However, with the reduction of the operating height of drones, the obstacles faced by drones are becoming more and more complex, including trees, power towers and buildings, which greatly limit the low-altitude operations of drones. Therefore, the ability of UAV perception and obstacle avoidance plays a very important role in the development of low-altitude UAVs in the future. Among them, the UAV's ability to perceive the environment is the key difficulty and focus of attention. [0003] At present, there are few researches on UAV obstacle percept...

AI Technical Summary

Problems solved by technology

[0004] At present, the research on UAV environmental perception schemes mostly adopts the method of multi-sensor fusion, such as millimeter-wave radar, ultrasonic radar and visual fusion perception. Large size, complex system structure and other defects make this method difficult to be practically applied

There are few studies on perception using only visual sensors. The main reasons are immature technology, large amount of calculation and difficulty in meeting real-time requirements, etc.

Images