Research On Extraction Algorithm Of Firefighter Physical Training Information Based On Deep Learning

Firefighters are the main force of fire emergency work in China.The complex and changeable working environment and arduous rescue tasks put forward higher requirements for the basic physical quality of firefighters.In order to ensure the safe and smooth development of fire emergency work and improve the scientificity and effectiveness of firefighters daily physical fitness training,combined with the actual needs of firefighters physical fitness training,this paper constructs a object detection and human posture estimation algorithm based on deep learning,and studies the automatic and intelligent extraction and analysis of relevant evaluation information in firefighters physical fitness training.The main research work of this paper is as follows:(1)Firefighter physical fitness training scene object detection and motion track analysis.Collect the firefighter physical fitness training image data,build the object detection data set through data enhancement,data annotation and annotation format conversion,and train the object detection model based on YOLOv4 network,so as to realize the detection of 11 types of targets of firefighters,wearing and carrying equipment in the scene of physical fitness training;On the basis of object detection,multi-object tracking algorithm is introduced to improve the effect of object detection,output the moving track of the target,and calculate the moving speed of firefighters according to the relationship between pixel distance and actual distance after image perspective transformation under fixed shooting angle.(2)Pose information extraction and physical movement analysis of firefighters.Based on the object detection algorithm and human pose estimation algorithm,the pixel coordinates of single or multiple firefighter human joint points in the physical fitness training scene are extracted.For the basic physical fitness training poses,the poses are classified and counted by combining the distance measurement representation vector and K-Nearest Neighbor algorithm,and then the two-dimensional node coordinates in pixel coordinate system and the depth information of corresponding nodes in depth data are used to construct a real-time dynamic and high-precision three-dimensional model of firefighters pose in camera coordinate system,and the distance measurement based on three-dimensional model is realized.(3)Construction of intelligent assistant system for firefighters physical fitness training.Based on the results of object detection and human pose estimation algorithm based on deep learning,and combined with the actual needs,an intelligent assistant system for firefighters physical fitness training is constructed.The system adopts C/S architecture,takes Qt as the graphical user interface development framework,and the front and back algorithms are written in C + + language.The system functions include six modules: initialization,registration and login,project management,data preview,data processing and data analysis.It can realize the visual processing of a series of core tasks such as object detection,target tracking and track analysis,single person and multi person pose estimation,physical pose classification and counting,pose three-dimensional model construction and measurement in the scene of firefighter physical fitness training.