基于YOLOv8目标检测模型的工地车辆清洁度识别

Cleanliness recognition of construction site vehicles based on the YOLOv8 object detection model

为缓解建筑工地车辆未经过清洗离开工地带来的城市道路污染和卫生问题,采用计算机视觉领域的目标检测方法即YOLOv8技术设计了一种车辆清洁度检测算法。该算法首先根据工地实际视频数据构建目标检测标签与清洁度标签,然后利用构建好的工地车辆清洗数据集训练YOLOv8目标检测模型,接着调用该模型用于实时检测目标车辆与清洗装置出水情况,最后根据检测效果实时计算最终清洗状况。针对清洗装置有时喷水面积较小的情况,在网络架构中额外引入一个专门用于小尺寸目标的检测层,并采用Focus层技术对输入图像执行分块处理,验证指标包括精确率、召回率、平均精度均值、帧率和清洁度(平均交并比)。结果显示该算法对车辆清洗情况的判断准确率达97.6%,表明所提出的改进模型很好地解决了工地车辆清洁度识别问题。

To alleviate the urban road pollution and hygiene problems caused by construction site vehicles leaving the site without being cleaned,a vehicle cleaning detection algorithm was designed using a target detection method in the field of computer vision,i.e.,YOLOv8 technology.The algorithm first constructed object detection labels and cleanliness labels based on actual video data from construction sites,and then used the constructed dataset for training the YOLOv8 object detection model specific to vehicle washing.The trained model was then employed for real-time detection of target vehicles and the water spraying situation of cleaning devices.Finally,the detected results were used to calculate the ultimate cleanliness status in real time.For the cleaning device sometimes with a small water spray area,an additional detection layer dedicated to small-sized targets was introduced into the network architecture,and the focus layer technology was used to perform block processing on the input image.Validation metrics included precision,recall,mean average precision,frames per second,and the final cleanliness level.The results showed that the algorithm achieved an accuracy of 97.6%in determining the vehicle washing situation,indicating that the proposed improved model effectively solved the problem of site vehicle cleanliness recognition.