主催: 一般社団法人 日本機械学会
会議名: ロボティクス・メカトロニクス 講演会2018
開催日: 2018/06/02 - 2018/06/05
For sea animal tracking survey, sensor-tag attaching method like bio-logging is widely used. As a new sea animal tracking strategy, we are developing the method which doesn't need to attach sensor-tag to animal by using AUV(Autonomous Underwater Vehicle) with multi-beam sonar. As the first step of animal tracking, we propose applying the YOLOv2[5], CNN-based object detection method, for sonar image to detect the animal and its position. To evaluate this method, we collected sea turtle data in a tank of Aqua World Ibaraki Prefectural Oarai Aquarium. After that, we labeled 541 sea turtle sonar images with bounding box. Fine-tuning with weights pre-trained on optical image dataset was used for training. Finally, this method could successfully detect the sea turtle in test dataset.