In order to find drowning person in swimming pool in time and reduce drowning mortality rate

by studying on Mask R-CNN algorithm

optimizing the convolution backbone of traditional Mask R-CNN algorithm

adding branches of cascade feature pyramid model

and designing drowning detection system of swimmer

through real-time recognition of swimmer’s posture in the swimming pool

determines the drowning person and raise alarm in time.Experimental result show that the detection speed of system is 5 fps

the detection rate is 93.3 %

and the false detection rate is 6.4 %.The effect is good and meets the expected requirements.