Design And Implementation Of Drowning Behavior Detection System In Swimming Pool Based On Mask R-CNN

Swimming is a popular sport,but the potential safety hazards can not be ignored.The traditional manual monitoring of swimming pools not only consumes manpower,but also is prone to omissions,resulting in loss of life and property.In recent years,computer vision technology has developed rapidly and has made remarkable achievements in many fields.The application of computer vision technology in swimming pool safety detection can replace the traditional manual surveillance method and provide safety guarantee for swimmers.In the aspect of using computer vision technology to detect swimming pool safety,although attention and research have been paid to it for many years at home and abroad,the traditional background subtraction method is often used in the early stage because of the limitation of related technical level.This method is only suitable for simple scenes,and in swimming pool scenes,the detection speed is slow,the recognition rate is low,and the detection effect is poor,so no substantial results have been achieved.In view of the above problems and challenges,this paper designs a swimming pool drowning detection system based on deep learning.The main work is as follows:(1)The basic information of common swimming pools has been investigated and the difficulties of drowning detection in swimming pool has been analysed.Aiming at the difficulties of background dynamics,reflectivity of water surface,morphological diversity of swimmers and diversity of objects in swimming pool scenes,an object detection method based on deep learning is adopted.And a classifier model with high robustness and weaker interference items is trained by training a large number of samples.(2)A drowning detection system for common swimming pools has been designed,which includes an information acquisition module,an object detection algorithm module and a graphic user interface module.In the information acquisition module,four high-resolution cameras are installed above the swimming pool,and multi-threading technology is used to transmit information synchronously;In the object detection module,Mask R-CNN algorithm is used to train a highly robust classifier model,which can detect swimming,up-right in water,drowning and standing on shore;In the graphic user interface module,the Qt Designer and PyQt have been used to draw and implement the interface,which the interaction is convenient,and also it is helpful to observe the swimmer's behavior in the swimming pool.(3)A comparative experiment was carried out to compare the overall performance of the classifier model mAP(mean Average Precision)under different iterations of Mask R-CNN algorithm.It was found that the higher the iterations,the better the performance of the classifier model.Finally,a higher number of iterations for the optimized Mask R-CNN classifier model has been used.The object detection experiments were carried out in four swimming pools using the above optimal classifier model.The results show that the detection speed of the system is 5fps,the precision is 91.2%,and the false positive rate is 7.5%,which meets the anticipated detection requirements of the system.