2021 Volume 77 Issue 2 Pages I_163-I_168
There are from 2,000 to 3,000 rescues every summer season on the beaches of Japan. At the time of drowning, a "help signal" is recommended worldwide as a rescue request sign. But it is not easy to find it because one life saver monitors over 1,000 beach users. On the other hand, in previous research, efforts have been made to automatically detect a help signal using AI. However, the video used for AI learning in the previous research was shot in winter, and there are no users other than the drowning victim (subject), and the subject is close to the camera and always faces the front to the camera. Therefore, we developed an AI that can detect help signals under multiple conditions where drowning may actually occur. Three AI models were developed using the object detection algorithm YOLO v3. As a result, the AI model which has most high accuracy was possible to detect help signals with 87 % F measure.