Research on Warehouse Object Detection for Mobile Robot Based on YOLOv5

摘要

In view of the lack of public datasets for object detection in warehouse environment, a large number of images containing cargos, trays and forklifts in real warehouse environment is collected and annotated by cameras, creating a warehouse object dataset. At the same time, in view of the low detection accuracy of traditional object detection algorithms in warehouse environment, the YOLOv5 series models are applied to warehouse environment, and the models are trained and optimized on the warehouse object dataset created by ourselves, achieving accurate detection of warehouse object. The mAP@50 of the YOLOv5 series models has reached more than 99%. Among them, the mAP@50-90 of the YOLOv5l model and YOLOv5x model is about 92%, which is better than other models. The YOLOv5m model is also deployed on the self-designed mobile robot, realizing the automatic detection of warehouse object. The experimental results show that YOLOv5m has achieved a satisfactory warehouse object detection effect, and it has certain practical value in the field of warehouse object detection.