Safety Helmet Detection Dynamic Model Based on the Critical Area Attention Mechanism

Abstract

In the substation monitoring environment, monitoring the helmet wearing of workers is an important approach to ensure safety. Because the helmet size in entire surveillance images is often small and the characteristic information is unclear, existing target detection algorithms present the problem of missed detection and omission. A safety helmet detection dynamic model based on the critical area attention mechanism first detected the human target in the image, and then, locked the human head area through the critical area attention mechanism network. Finally, the feature map of the critical areas of the head was up-sampled many times to increase the proportion of the helmet area in the image to highlight the characteristic information of the helmet in the image. The algorithm used the dynamic model method to match the optimum up-sampling times for the helmets of different scales, which improved the recognition speed of the algorithm while ensuring the recognition accuracy. The experimental results showed that the recognition rate of algorithm for helmets was 92.68%, which is considerably higher than that of other target detection algorithms in the same field.