Study of Taiwanese White Dolphins Detection and Tracking Techniques by utilizing Autonomous Unmanned Aerial Vehicles
Journal
Oceans Conference Record (IEEE)
Journal Volume
2022-October
ISBN
9781665468091
Date Issued
2022-01-01
Author(s)
Abstract
The paper describes the autonomous virtual UAV (Unmanned Aerial Vehicle) using a visual detection model to track the Taiwanese white dolphins. In this research, the SITL simulator (Software in the Loop) runs ArduPilot on the machine. ArduPilot is an autopilot system for autonomous vehicles, which is installed on the virtual drone. The simulation helps evaluate the detection model and the tracking system prior we launch the drone out over the ocean where the consequences of mistakes can be very expensive. The customized YOLO v4 model has been trained on the Taiwanese White Dolphin Image dataset 201S owned by Cetacean Lab in NTU. The model has the ability to detect the side view of the white dolphin under 702 pixels@0.5AP (Average Precision) in the real-sea footage. The model's proper detection condition such as the distance between the white dolphin and the drone, the height above the horizon, and the pitching angle of the camera, is estimated by the simulation in the paper as well. A white dolphin 3D model is sculpted and textured based on the real white dolphin called Doufu with Blender software. In the Gazebo Simulator, the 3D dolphin model moves in the shapes of a rectangle, circle, and figure-eight with a constant velocity of 10 knots on the water. With the frame caught on the virtual drone camera and the detection model, the virtual drone can track the dolphin without losing it under certain conditions. The Taiwanese White dolphin (subspecies of Sousa chinensis) is listed as critically endangered [1]. And the population is decreasing year by year. The real-time monitoring of the Taiwanese White dolphin is still unlikely to be realized unless we launch the drones, send the hydrophones to the sea, and establish the individual PHOTO ID and AUDIO ID. It is hoped that people can take advantage of this system, run different varieties of simulations, and Figure out a better algorithm running through those possibilities to help monitor and conserve the white dolphins.
Subjects
DNN | MOOS-IvP | Realtime video object detection | Taiwanese White Dolphin | Unmanned Aerial Vehicle | Unmanned Vehicle | YOLO v4
Type
conference paper