Research And Application Of Hard Hat Detection And Tracking Algorithm Based On Deep Learning

In economic construction,the importance of construction production is self-evident,but production safety is also crucial.At the construction site,safety helmets,as the last line of defense for workers’ lives,bear a crucial responsibility.To ensure construction safety,it is not only necessary to encourage workers to consciously comply with safety regulations and wear safety helmets,but also to establish efficient and accurate monitoring mechanisms.Although existing deep learning algorithms can be used to detect whether workers on construction sites are wearing safety helmets,there are still some problems,such as easy omission and misdetection of low resolution images and small targets,repeated recognition of the same target,which can increase the time complexity of algorithm operation and affect the accuracy of the algorithm.Based on the above issues,this article designs and builds a safety helmet wearing detection application system based on deep learning.The main work of this article is as follows:(1)Adopting the improved YOLOv5 mainstream framework for helmet wearing detection.Firstly,to address the issues of missed and false detections in low resolution images and small targets in the YOLOv5 detection algorithm,a new CNN building block called SPD-COnv was adopted to replace each convolution step and each pooling layer,effectively improving the model’s detection ability for small targets and low resolution images;Secondly,by adding an attention mechanism SE to adaptively adjust the correlation and weight between channels,the model enhances the extraction and filtering of key information;This article collected and annotated a total of 6000 images to form a dataset that conforms to the experiment in this article.The dataset was used for training and validation,and the experimental results showed that the improved model has a good improvement in detection accuracy.(2)Using an improved DeepSort algorithm for target tracking.Firstly,to address the issue of IOU being unable to measure the adjacent distance and overlap between the prediction box and the tracking box,GIOU is used instead of IOU to improve the matching ability of the DeepSORT tracking algorithm;Secondly,in view of the poor ability of the original algorithm to extract the appearance features of the target,we combined Darknet53 with the DeepSort algorithm to conduct down sampling to better extract the Semantic information and visual features of the target,so as to improve the performance of the model in the target tracking task;Finally,the improved DeepSORT algorithm is combined with the detection algorithm to improve the accuracy of the entire model.(3)Designed and implemented a helmet wearing detection system,and applied the trained model to the system.The system can provide users with functions such as image detection,video detection,camera detection and tracking,and alarm.After testing,the system has shown good response speed and recognition accuracy.