Research On Road Obstacle Detection And Ranging Algorithm Based On YOLOv5

With the large-scale application of intelligent driving technology,vehicles with auxiliary driving functions are rapidly developing.The perception of road obstacles is the key to whether vehicles can avoid obstacles in a timely and effective manner.Accurately measuring the distance between vehicles and obstacles is an important guarantee for safe driving.Therefore,obstacle detection and ranging are research hotspots in intelligent driving technology.This article proposes a road obstacle detection and ranging algorithm based on YOLOv5,and through actual vehicle verification,it can quickly and accurately detect obstacles,obtain the distance between vehicles and obstacles,and provide technical support for intelligent and safe driving.The main work content is as follows:(1)Network and offline collection of pavement obstacles for pedestrians and traffic cones,and datasets are made using Label Img software.The network structure of YOLOv5 algorithm is improved,CA and ECA modules are added to the backbone structure of YOLOv5,and small target detection layer and corresponding feature extraction layer are added at the output end and head of the network.Through the ablation experiment,the average detection accuracy and recall of YOLOv5 before and after improvement were compared and analyzed.(2)A monocular camera is used to measure the distance to the target obstacle.Firstly,the internal and external parameters of the selected camera are calibrated,a monocular ranging model introducing side declination angle and pitch angle is established,a human-computer interaction interface is established in the Py Qt5 software environment,and image and video detection are realized by designing corresponding functions.In order to make the measurement data as accurate as possible,in the actual measurement,three points with a lateral interval of one meter are selected,and their longitudinal distances are measured and averaged respectively,which verifies the ranging accuracy of the model for road obstacles.(3)The improved YOLOv5 algorithm model is transplanted to the domestic Jetson Nano 4G development board for edge deployment to realize real-time detection and ranging of vehicles to pedestrians and traffic cone.Based on the Deep Stream framework,establish a detection environment for Source/Deep Stream Yolo master.By calling the ranging model algorithm and setting relevant parameters,real-time detection can be achieved.Conduct actual vehicle loading experiments on the road to verify the accuracy and effectiveness of the improved YOLOv5 algorithm detection.