Swimmer motion analysis with application to drowning detection

We present a novel swimmer motion analysis approach for detecting possible drowning incidents in a swimming pool. The drowning incidents are detected by examining the sequence of motion and shape features extracted from swimmers in the pools. Incorporated with the knowledge of professional practice in water crisis recognition, two event-inference modules have been developed: one evaluates the condition of a swimmer using a set of reasoning rules and another relies on hidden Markov models (HMMs) to recognize drowning behavioral signs. Both modules have been applied to a number of video clips of simulated drowning incidents and promising results have been obtained.