Detection Of Abnormal Human Behavior In Swimming Pools Based On Multi-source Information Fusion

With the development of society and the improvement of people’s living standards,fitness has become an important topic of concern,and the demand for swimming pool is also growing.Therefore,the standardized management of swimming pool,especially the prevention of dangerous situations,has become increasingly important.In the construction of smart swimming pool,prevention of drowning,prohibition of smoking,avoidance of accidental falls,prevention of fights and other abnormal behaviors are all important considerations.In this thesis,drowning and smoking in swimming pool were studied.Currently,most research on abnormal human behavior is monitored using visible light videos.Although visible light videos have clear texture information and high spatial resolution,they cannot accurately identify whether swimmers are above or below the water surface.In addition,due to factors such as water surface reflection and human body diversity,there can be false positives and false negatives in person identification.In contrast,thermal imaging technology relies on the thermal radiation of objects and has the advantage of weak penetration into water,it can distinguish whether a person is above or below the water surface and can detect cigarette butts with higher temperatures.UWB(Ultra Wide Band)tag identification,on the other hand,is unique and can quickly and accurately identify the identity of swimmers through UWB,with precise positioning and tracking capabilities.Therefore,by combining these three technologies,they complement each other and can automatically identify and detect abnormal drowning and smoking behavior in swimming pool scenarios.Based on this,this thesis carried out the following work:(1)Designed target detection algorithms for the three modes of visible light video,thermal imaging video,and UWB.For visible light videos,a distortion correction algorithm based on checkerboard images was proposed.After correcting the distorted images in the swimming pool,YOLOv5 and Deep Sort were used for target detection and tracking.For thermal imaging videos,a target detection algorithm based on mean background difference modeling was proposed.For UWB technology,a UWB positioning scheme was designed to determine the position of personnel in the swimming pool,and the accuracy and latency of UWB positioning were tested.(2)Designed a multi-modal target identity consistency matching algorithm.When UWB,visible light video,and thermal imaging video capture the location information of the target personnel,the location information of the three modes of the same person needs to be accurately matched.Therefore,a global coordinate system was established for the swimming pool,and corresponding transformation relationships were designed for the coordinate systems under the three modes to obtain their position information in the same global coordinate system.Finally,identity matching was performed according to the matching rules.(3)Designed a drowning detection algorithm based on visible light video,thermal imaging video,and UWB.A human abnormal behavior detection algorithm based on behavior parameter fields was proposed.The algorithm extracts the features of normal swimming for each swimmer and establishes a behavior parameter field to express the regularity of human behavior.By comparing regular behavior,the abnormal behavior coefficient is calculated.Combining the features of swimmers under the three modes of UWB,visible light video,and thermal imaging video,drowning rules were set to judge normal swimming and drowning behavior of swimmers.(4)Designed a smoking behavior detection algorithm based on visible light video and thermal imaging video.The algorithm extracts human body and facial keypoints from each frame of the visible light video,and establishes a smoking action discrimination model based on human behavior features,effectively identifying smoking actions.In addition,in terms of thermal imaging video,the thermal radiation imaging of the object is used to detect the presence of cigarettes,and the smoking behavior is judged by synthesizing information from both human body movements and cigarette detection.(5)Based on the above content,a system for detecting abnormal human behavior in swimming pools has been designed.In terms of hardware,the installation and testing of devices such as UWB,visible light cameras,and thermal imaging cameras have been completed.In terms of software,the development of the swimming pool human abnormal behavior detection system has been completed,including the member entrance and exit subsystem and the abnormal behavior detection subsystem.The system integrates various algorithms and realizes real-time detection of drowning and smoking behavior in swimming pools.