Abstract
This report is discussing the "state of art" in automated video analysis that could be used
for automated real-time analysis of fish behaviour in RAS facilities, the main objective of
WP4 in DigiRAS. Two systems are discussed in this report. (i) A 2D fish tracking system
based on a mono camera system where Faster R-CNN or YOLO is used for detection and
object pairing and Norfair is used for object tracking. (ii) A 3D fish tracking system based
on a stereo camera system where Stereo R-CNN is used for detection and object pairing,
OpenCV is used for stereo matching and Norfair is used for object tracking. A set of
formulas is suggested for calculating fish swimming speed, fish swimming pattern and
population synchronization from the fish tracking algorithms, these values could be used
to get a general picture of the behaviour of the fish population in the tank.
for automated real-time analysis of fish behaviour in RAS facilities, the main objective of
WP4 in DigiRAS. Two systems are discussed in this report. (i) A 2D fish tracking system
based on a mono camera system where Faster R-CNN or YOLO is used for detection and
object pairing and Norfair is used for object tracking. (ii) A 3D fish tracking system based
on a stereo camera system where Stereo R-CNN is used for detection and object pairing,
OpenCV is used for stereo matching and Norfair is used for object tracking. A set of
formulas is suggested for calculating fish swimming speed, fish swimming pattern and
population synchronization from the fish tracking algorithms, these values could be used
to get a general picture of the behaviour of the fish population in the tank.
