Detection Of Drowning Behavior Based On Skeleton Trajectories

At present,video-based drowning detection algorithms in this field ignore the local features of the human body when drowning occurs,such as the waving of the human arm and the floating head below the water surface when drowning,and so on,most only take into account the overall characteristics of the human body.In order to quickly and accurately identify dangerous behavior in swimming pools,based on real drowning video,a fast and real-time swimming pool drowning detection algorithm based on human skeleton information is proposed in this paper.The main work of this paper includes the following four aspects:(1)using the distance between the skeleton to remove the repeated recognition of skeleton.(2)the Hungarian algorithm and KM algorithm are used to correlate and match the skeleton information of the front and back frames to identify the skeleton information of the same person in the front and back frames.(3)using Kalman filter to predict and update skeleton.(4)based on real drowning video,the detection rate of skeleton and velocity variance of head skeleton are synthetically analyzed to judge human behavior and identify drowning behavior.Finally,the experimental results show that the algorithm has good stability and accuracy in judging the swimmer’s state,and has advantages in identifying the dangerous behavior in swimming pool.Alphasse is an excellent human posture estimation framework,which can extract the skeletal information of human body more accurately.And from the human skeleton trajectory can extract more specific local features of the human body to the swimmer’s behavior for a more accurate judgment.