a. Midwater Organism Tracker
PICASSO will semi-automatically track animals in midwater. In order to detect and track an animal the vehicle has to incorporate animal image recognition and then automatically move so as not to lose the animal that has been recognized. JAMSTEC has developed a prototype system for an animal tracker using the MROV vehicle. To simplify the prototype system, only the pan-tilt system of the camera rather than the entire vehicle itself was controlled by the tracking program. Color deference in HLS (Hue, Saturation, Luminance) color space was basically used for detection. Identification of a target is initially done by clicking on the target on the display. RGB values in the 9 x 9 pixels around the pixel clicked are converted to the HLS color space. A center of gravity for pixels with the near-HLS value obtained is then calculated. When the distance between the center of gravity and the center of the image obtained by the camera exceeds a preset limit of the pan-tilt, the program controls the camera to center the animal in the middle of the observation space. The program also has a displacement prediction function to predict movements of animals. A detection and tracking test was carried out in the large fish tank (6.5 m in depth, 144 m2 area of base) at the Enoshima Aquarium. A scene taken during the test is shown in Figure 13. The prototype system was able to detect and track a small fish (red circle in the figure) for 30 seconds in this test. However, in most cases the duration of capturing the target was only a few seconds because there were many fish in the tank and the background-target contrast was low compared to in the midwater zone of the ocean. For more accurate detection, they will collaborate on an image recognition method with MBARI (Walther 2004). This method simulates human vision functions and has a high target recognition probability. JAMSTEC will also investigate a program to track animals by linking this output with thruster control.
Was this article helpful?