The Swimmers Detection And Tracking Based On Video Monitoring

Swimming is a very popular sport,it has a lot of benefits such as bodybuilding,weight loss,resistance and cardiac function enhancement and so on.But coming with the problem of drowning which also causing troubles to the majority of the people.In China,the number of pool lifeguards fails to meet the demand,and a person's energy is limited,only rely on the lifeguard can not meet the demand of real-time drowning rescue.Thus,a recent study on the pool of intelligent video monitoring system has become a hot direction to researchers at home and abroad.This paper focused on the swimmers detection and tracking on the pool environment.While the existing detection and tracking algorithms are relatively mature,but still there exist no algorithm for object detection and tracking of any environment can achieve good results.Achieving swimmers detection and tracking is the pre-condition of analysing behaviors and judging whether he or she is drowning.Because of the special nature of the environment such as water fluctuations,the lane dividers and the gaps between the tiles waggle,etc.,the pool has become a very complex dynamic background.Therefore,the swimmers detection and tracking have become a difficult problem.This paper studied from the video pre-processing,swimmers detection,swimmers tracking and drowning behaviors analysis four aspects.Works were done as follows:(1)Video pre-processing: first analyzed the characteristics of the collected pool video images.While the buoy movement,the lane dividers and the gaps between the tiles waggle which caused by the water fluctuation,made the pool a very complex dynamic background.In this paper,the pool image preprocessing methods based on the Mean Shift algorithm was used to reduce noises,which achieved good results in reducing the noises caused by the water fluctuation.(2)Swimmers detection: The results of three different background subtraction algorithms(GMM,Codebook,Vibe)were compared,wherein Vibe did the best in the foreground detection.And a swimmer detection algorithm based on the improved Vibe was proposed to improve the Vibe algorithm in the background model initialization and the swimmers detection two aspects.In terms of background model initialization,aiming at the problem that the Vibe algorithm uses the first frame to build the background model tends to occur ghost phenomenon,an improved Vibe background model initialization algorithm which combined with the median filtering algorithm was proposed.An image which contained a small amount of foreground was taken by done the median filter with the sampled video frames pixel by pixel.In swimmers detection,according to the characteristics of the swimmers,a series of specific treatments were done on the binary image to determine the position of the swimmers.(3)Swimmers tracking: the commonly used on moving object tracking algorithms were briefly introduced.In this paper,Particle Filter was used to track swimmers.Firstly,the theory of the Particle Filter algorithms based on the color distribution model was introduced,which achieving the tracking of single swimmer that was selected by mouse.Secondly,an algorithm based on color distribution model of Particle Filter on the multi-swimmers tracking data association was proposed,which realizing multi-swimmers tracking.The experiments results and analysises were given.(4)Drowning behaviors analysis: the characteristics of drowning behaviors were analyzed,and according to the captured video,several features used in drowning incident detection were assumed.Finally,the state transition diagram for drowning incident detection and alarms is designed.