Intelligent Safety Monitoring and Early Warning System for Construction Site

Faced with the complex environment and difficult construction of infrastructure projects, designing an intelligent safety monitoring and early warning system for construction sites can effectively detect existing violations and reduce the probability of accidents. Existing violation detection methods for construction sites mainly include hand-crafted feature extraction and deep neural network. However, the method of extracting features is usually difficult to design and the architecture of the deep learning-based method is simple, which might lead to poor detection performance in extreme cases (Insufficient light, small detection object, occlusion, etc.) and cannot be used in actual detection environment. Therefore, we improve the existing target detection algorithm by adding image preprocessing module, multi-scale feature fusion module, and repulsion loss term. We also use the KCF algorithm to continuously track targets to identify specific violations. On this basis, we develop an intelligent safety monitoring and early warning system to classify the detected violations and send the information to the responsible person in time, which significantly improves the management capacity at the construction site. Through a series of experiments, we compared the impact of different modules on detection accuracy. The results show that our model has a significant improvement compared to existing methods on our dataset, especially in harsh environments.