Detection of Key Points for a Rainbow Trout in Underwater Video Surveillance System

Abstract

Video Surveillance Systems (VSS) are progressing towards the industry digitalization. In this demo, we apply the Artificial Intelligence (AI) technology to detection of key points for a rainbow trout. The well-known YOLO-Pose neural network is typically used for human pose recognition in video recordings. We use this neural network for detection of 8 key point for an individual fish. Our constructed dataset includes 10 500 images for approximately 30 thousand fishes. The YOLO-Pose network is trained on the dataset using our combined loss function. Our early experiment shows that that detection works fine even in muddy underwater conditions.

Keywords

• aquaculture digitalization
• image recognition
• video surveillance system
• key points detection