基于YOLOv5算法的道路场景目标检测应用研究

Research on Application of Road Target Detection Based on YOLOv5 Algorithm

道路目标检测与识别在当前自动驾驶领域具有重要的意义,而在道路目标检测与识别中,检测算法的高精度和快速推理速度对于安全的自动驾驶至关重要。运用YOLOv5(YOLO,youonlylookonce)目标检测算法可以对车辆行驶途中的道路目标进行识别与分析,起到辅助驾驶并且降低安全隐患的作用。通过使用改进的YOLOv5目标检测算法训练BDD100K数据集,所获模型可以显著提高召回率,从而提高准确性。改进的YOLOv5算法主要是使用K-means算法来寻找数据集的最合适的anchors,并且通过实时数据增广训练得到更加精确的模型。结果表明,在BDD100K测试集上,改进后模型的mAP-50可以达到51.8%,并且,相比于原始模型的性能,改进后的模型对目标检测的mAP获得了明显的提高。相比于人工,本文模型可以在保证检测速度的情况下,更精确地检测目标。

Road target detection and recognition are of great significance in the current field of automatic driving.In road target de⁃tection and recognition,it’s essential for safe automatic driving that the detection algorithm is high-precision and makes an inference very fast.The YOLOv5 target detection algorithm can be used to recognize and analyze the targets of road,which can assist driving and reduce safety hazards.In this paper,by using the improved YOLOv5 target detection algorithm to train the BDD100K dataset,the ob⁃tained model can significantly increase the recall rate,thereby improving the accuracy.The main method used in this paper is to use the K-means algorithm to find the most suitable anchors of the BDD100K dataset;and by real-time data augmentation during training to obtain a more accurate model.The result shows that on the BDD100K test set,the mAP-50 of the improved model can reach 51.8%,and compared with the performance of the original model,the improved model has a significant improvement in the mAP of object de⁃tection.Compared with manual work,the model in this paper can detect targets more accurately while ensuring the detection speed.