Portable Road Condition Recognition System Based on Yolov5

In order to solve the social problems of the large base of visually impaired people in China, the high construction cost of visually impaired facilities, and the safe travel of visually impaired people, a portable basic road condition recognition system was designed. Road condition recognition relies on convolutional neural networks, dynamic recognition model training, and the use of multi module algorithms to achieve dynamic feedback of road conditions. The hardware design is based on a raspberry pie and a camera. Images are uploaded to the cloud in real time through the camera. The program interprets its data structure and describes the results. In the training phase of the road condition recognition model, using adaptive anchor frame calculation and adaptive image zooming can accurately frame vehicles and pedestrians. The road condition recognition model recognizes and displays the color of traffic lights through binarization of images, extracts text information from road signs, roughly measures the distance between objects and users through monocular ranging, and generates corresponding voice announcements.