基于YOLOv5的锥桶识别技术研究与应用

Research and Application on Cone BucketIdentification Technology Based on YOLOv5

基于计算机视觉理论与目标检测算法,利用YOLOv5模型和自制数据集实现对锥桶的识别。然后将训练好的权重部署到ROS智能小车上实现了小车自动驾驶中的自主避障功能。实验数据表明,本文仅仅利用95个图片,514个标记经过50轮训练就实现了97.36%mAP@0.5,对锥桶的识别效果很好,且具有较强的泛化能力。该锥桶识别模型有效提高了在复杂光学场景及密集锥桶目标下的识别准确率。

Based on computer vision theory and object detection algorithm,the cone buckets was realized by using YOLOv5 model and homemade datasets.Then the trained weight was deployed to the ROS smart car to realize the autonomous obstacle avoidance function in the car autonomous driving.Experimental data showed that this paper achieved 97.36%mAP@0.5 by using only 95 pictures and 514 markers after 50 epochs of training,which is very good in the recognition of cone buckets,and has strong generalization ability.This cone buckets recognition model effectively improves the recognition accuracy in complex optical scene and dense cone buckets target.