Based on Improved Single Shot MultiBox Detector Construction Site Helmet Detection Algorithm

For the existing engineering construction now the worker's helmet wear detection method consumes labor and time and also easy miss detection problems, to protect the safety of construction site personnel, proposed an improved Single Shot MultiBox Detector(SSD) helmet wear detection algorithm. First, the backbone network with Resnet-50 model instead of VGG-16 is used as the SSD algorithm, and the residual structure in Resnet can improve the feature extraction ability of the network; CA (Coordinate Attention) module is added before the feature layer enters the prediction to enhance the capture of localization information of the target. The experimental results show that the average accuracy (mAP) of the improved algorithm can reach 94.5% on the homemade helmet data set, which is 4.49 percentage points higher than the original algorithm, and can meet the accuracy requirements of helmet-wearing detection under construction sites.