Detection of helmet use among construction workers via helmet-head region matching and state tracking

Accidents at construction sites are prevalent, posing a significant safety threat to workers. Helmets play a crucial role in protecting workers' heads during accidents, and helmet wearing monitoring is essential for ensuring workers' safety. However, it becomes challenging to detect whether workers are wearing helmets when their heads are obstructed or invisible. To enable continuous and accurate monitoring of workers' helmet-wearing states, this paper proposes a method based on the YOLOv9 object detection algorithm, the YoloPose human pose estimation model, and the StrongSORT tracking algorithm for helmet-wearing detection. The keypoints detected by YoloPose are used to extract the head region and are subsequently matched with a helmet bounding box detected by YOLOv9. Based on the tracking results of workers, matching information from preceding frames is integrated to update workers' helmet-wearing states. The proposed algorithm achieves 98.89 % accuracy on the self-built dataset, significantly enhancing the consistency of helmet-wearing state detection.