Model for Recognition of Personal Protective Equipment in Construction Applying YOLO-v5 and YOLO-v7

Due to the nature of activities carried out during a construction, companies find it necessary to stablish controls to avoid accidents caused by the lack of personal safety equipment. However, this process often involves manual supervision carried out by a specific staff. Due to this, there is no efficient control, nor constant monitoring to prevent this type of accident. For this reason, a model is proposed that recognizes the use of personal protective equipment in construction through 2 deep learning models: YOLO-v5 and YOLO-v7. Four stages were carried out: (i) data collection, (ii) data processing, (iii) training and (iv) evaluation of results. A dataset of 743 images composed of 6 classes was used: boots, glasses, gloves, helmet, person, and safety vest. The results show that the Yolo v7 model obtains a 0.873, while Yolo v5 obtains a 0.796. These values represent the combination of precision and recall in mAP@0.5, therefore v7 was more optimal than Yolo v5. Additionally, the study shows which is the recognition pattern for both algorithms, this pattern helps us to reinforce the detection of objects through training.