Two-stage helmet security detection model combining CNN and MTCNN algorithm

Abstract

Physical safety is of utmost importance for workers in an industrial and construction environment. Fatalities related to the violation of wearing a safety helmet account for a large proportion of occupational deaths. Hence, the requirement of a high-efficiency automatic system to monitor and promote wearing helmets is of vital importance, reducing hours from manual monitoring, facilitating efficient education and enforcement campaigns that increase industrial safety. In this paper, a combined Convolutional Neural Network (CNN) algorithm and Multi-Task CNN (MTCNN) to monitor violations of safety helmets is proposed. All images are sourced from industrial and construction locations, undergoing various image processing techniques before model detection. The proposed model includes the MTCNN locating individuals' heads from the picture data, followed by a CNN assessment to determine helmet wearing. Under the same evaluation criterion, the result, where an obvious detection accuracy improvement is shown, illustrates the effectiveness of the proposed model in the comparative study. Therefore, the research demonstrates the effectiveness of incorporating multi-scale feature extraction techniques for enhancing model performance. This establishes the groundwork for future research within this domain, facilitating the creation of more advanced helmet detection or other relevant items recognition systems that can be utilized in real-world applications.