Research On The Underwater Alarming Algorithm Based On Visual Analysis

With the development of swimming sports,the requirement of underwater lifesaving is increasing greatly.It doesn't satisfy the need of real-time drowning lifesaving if it only relies on the lifeguards.In recent years,underwater video surveillance system has become the hot spot of widespread attention at home and abroad.The biggest problem is the algorithm research of visual analysis of human motion.According to the implementation requirements of underwater video surveillance system,and through analyzing a variety of existing image processing algorithms,this paper explores the three parts of the algorithm systematically:moving human target detection,tracking and behavior understanding.On the basis of the characteristics of video images,two types of human target detection algorithm are put forward in this thesis:(1) The algorithm of gray image detection based on area.Firstly,it adopts the blocking operation due to the slow running speed of high-resolution image and the much noise in swimming underwater.Then, texture feature is introduced to solve the problem of human shadow after extracting statistic feature of gray histogram.(2) The algorithm of color image detection based on pixels.It firstly uses statistic inference method to segment moving target,and then applies morphological operation to eliminate noise,finally get the regions of moving targets by the connected region operation.Considering the moving information of human targets,Kalman filter is used for object tracking in this thesis.After introducing the Kalman filter principle and its application in tracking in detail,the paper builds the moving model and setups relevant parameters.As a result,the target tracking is completed well.At the same time,it also conducts the feasibility analysis for the application of Nearest Neighbor method in multiple moving objects tracking.Finally,by analyzing the human moving patterns under water,a motion understanding method is presented which is based on finite state machine.This method can distinguish between normal swimming and drowning effectively,so drowning alarming is realized fast and accurately.