Research on helmet detection algorithm based on android system
Zheng Xie, Lantao Su, ZhenHuan Zhang, Hong Biao · 2023
Aiming at the demand of real-time detection of site helmet wearing, an artificial intelligence application based on android embedded system is designed to promote the accurate management of smart site safety. The dataset is obtained by Internet crawlers and construction site videos. Samples are divided into training set, validation set and test set in the ratio of 8:1:2. The yolov5s is used as the objective detection algorithm, and the network to be modified. After training the model is quantified, and then the model calls the interface and the lightly quantified model to improve the inference speed of the model. The accuracy of the trained model inference is 92%, the mAp is 80.4%, and the detection speed is 38s. After quantization and deployment to android system, the accuracy is 89.2 and the mAp is 87.6%. The detection speed is 19.2ms, which maximizes the detection speed under the premise of ensuring accuracy. The design helps to detect the safety helmet worn by site personnel automatically and promote the safety management of personnel at smart sites.