Research On Unsafe Behavior Detection Devices In Construction Sites

The increasing development of the housing construction market has brought about the frequent occurrence of safety accidents.The hidden dangers of the housing construction site are caused by the unsafe behavior of construction personnel.It is increasingly important to strengthen the effective supervision of the unsafe behavior of construction personnel in the construction industry.The current prevalent detection method is by the supervisory staff with a manual approach to occasional safety inspections,but the method is inefficient and lacks real-time,so it is of great significance to study a human-independent method of detecting unsafe behavior.This paper proposes an unsafe behavior detection device for housing construction sites,which takes Jetson nano as the carrier,takes housing construction sites as the research background,and combines with deep learning technology to establish an unsafe behavior detection model based on the improved YOLOv5 s algorithm,and realizes automatic unsafe behavior detection for housing construction sites.A new detection network structure is designed to fuse the first CSP structure in the backbone network with the large-size detection feature map after down-sampling processing to obtain a new multi-scale feature fusion detection network structure in order to enhance the detection capability of the model on small targets;meanwhile,the Focal loss method is introduced in the training to improve the confidence formula of positive and negative samples in the loss function,which effectively improves the The model has an unbalanced number of positive and negative samples in the training process;the unsafe behavior dataset is established by collecting images online,and the dataset is clustered using the Kmeans++ clustering method to obtain the anchor frame size that meets the target dataset,and the unsafe behavior detection model is obtained by combining the above work;then the model is quantified and accelerated based on the TensorRT framework and deployed on the Jetson nano edge computing device.Based on these works,we completed the design and deployment of the unsafe behavior detection system model;then we completed the image real-time transmission function based on GStreamer technology,and completed the Web-side function design by using cloud server technology and Apache framework to realize the data upload and data access and online alert function on the Web side of the cloud server.Finally,the design of an unsafe behavior detection system for housing construction sites was completed.The results of several sets of experiments show that the overall operation of the housing construction site unsafe behavior detection device designed in this paper is smooth,the unsafe behavior detection model obtained based on the improved YOLOv5 s algorithm performs better than the original algorithm on the homemade data set,and the quantified accelerated model can better complete the task of automatic detection of unsafe behavior on housing construction sites in the Nano edge computing device.In addition,the device has some reference value for the subsequent implementation of a more complete and practical site unsafe behavior detection system.