Detection of mobile phone use by labourers on construction sites using yolov11-s

Abstract

Mobile phone distractions among construction labourers pose significant productivity challenges. This study presents a YOLOv11-s-based model to detect and classify construction labourers who use mobile phones during work. The system was trained using a person dataset, a helmet dataset and a mobile phone dataset, obtained from an online database and custom images collected from Sri Lankan construction sites. The proposed system followed a four-stage approach, beginning with person detection, followed by helmet detection and classification. Then, through image preprocessing, the model analysed the helmet colour using histogram analysis and the Hue Saturation Value colour scale to detect labourers with yellow helmets. Subsequently, the performance evaluation metrics, such as precision-recall curve, mAP@0.5 and inference time, indicate that the trained model performs better on the testing data in detecting construction labourers who are using mobile phones during work. Finally, mobile phone detection is carried out. Images from Sri Lankan construction sites were used for deployment validation and to check for model overfitting. The system can be further developed by using motion detection through IoT to detect the continuous use of mobile phones through timeframe analysis. This study contributes to improving workplace productivity through the automated detection of distractions in construction.