基于YOLOv5与RGB-D融合的交通锥桶识别与定位方法研究OA

To meet the requirements of real-time traffic cone detection and three-dimensional localization in construction and testing scenarios involving traffic cones,and to alleviate false positives and missed detections under complex illumination and occlusion conditions,this paper proposed a traffic cone recognition and localization method based on YOLOv5 and RGB-D fusion.First,863 cone images featuring diverse lighting,viewing angles,and backgrounds were collected and annotated,and divided into training and validation sets in an 8∶2 ratio.Through data augmentation,hyperparameter tuning,and an early stopping strategy,the optimal model weights were obtained.Next,the trained model was deployed on an Intel RealSense D435 camera,achieving synchronized acquisition and spatial alignment of RGB and depth frames.Finally,depth information was extracted from the center region of the detection bounding box and combined with camera intrinsic parameters for back-projection to estimate 3D coordinates.Experimental results indicate that the proposed method has high detection accuracy(Precision≈99.8%,Recall≈99.6%,and mAP@0.5≈99.5%),with a processing speed of approximately 26 FPS(≈37 ms per frame),an average localization error of less than 5 cm.The method maintains stable performance under challenging conditions such as strong glare,shadows,and partial occlusion.This method integrates 2D detection and 3D localization of traffic cones,ensuring high detection accuracy while balancing real-time performance and localization precision.It can provide reference for traffic cone perception.