【摘要】 在人工智能蓬勃发展的新时代,面对传统障碍物硬件设备检测成本高、实时性差、无法常态化进行、依赖于人工控制、耗时耗力等不足,设计了基于深度学习的目标检测方法。而在实际情况中,路面经常会出现光照不足、大气杂质、光学系统失真等恶劣环境,从而导致拍摄图像模糊不清晰,造成图像的严重退化,极大影响了后续路面障碍物的管理工作。本文将从此角度出发,在以往的道路障碍物识别方法上进行改进,利用Retinex理论在YOLO基础网络上进行改进实现恶劣环境下的路面障碍物检测,从而实现道路路面的自动化检测,提高路面管理部门的工作效率。更多还原

【Abstract】 In the new era of the vigorous development of artificial intelligence, in the face of the problems of high cost, low real-time, unable to be normalized, relying on manual control, time-consuming and labor-consuming of traditional obstacle hardware equipment detection, a target detection method based on deep learning has come out.However, in the actual situation, the road often appears poor lighting, atmospheric impurities, optical system distortion and other harsh environment, which leads to blurred images, resulting in serious degradation of images, greatly affecting the subsequent management of road obstacles.This paper will start from this point of view, improve the previous road obstacle recognition methods, and use Retinex theory to identify road obstacles.In order to realize the automatic detection of road surface and improve the work efficiency of road management department, the improvement of YOLO based on network is carried out to realize the detection of road surface obstacles in bad environment.更多还原