Automated Optimised Segmentation of Swimming Fish Midlines


Student thesis: Master's ThesisMaster of Philosophy

Original languageEnglish
Awarding Institution
Award date2021
Show download statistics
View graph of relations


While many fish propel themselves using their continuous flexible bodies, fish robots and biological models are often constructed from interconnected discrete segments. Considering this, how many segments are required for a robot or model to represent fish movements accurately? To bridge the gap between biology and engineering, two new methods are presented here, that automatically determine accurate and concise segmented fish models from actual fish data. These two methods are: segment growing approach, a greedy algorithm that sequentially 'grows' segments across the fish midline, and evolutionary algorithm approach, which progressively 'evolves' different combinations of segment lengths. These methods identify key bending points along the fish body, linking them with rigid segments, so that the difference between fish and modelled midlines is minimised. To verify the utility of these methods, they were tested with the kinematics of ten species during steady swimming, and rainbow trout over four swimming behaviours, and multiple swimming speeds. Broadly categorised as (sub)carangiform swimmers, these fishes exhibit diverse morphology, along with various swimming patterns. From these tests, several trends in results are found: 1) Regarding segment numbers, five segments are sufficient for modelling the 23 kinematics of all fishes with at least 99% accuracy (midline-segment difference < 0.01 body lengths). 2) Segment lengths get progressively shorter towards the tail, for multi-segment models with best performance. 3) Between different species, there is notable variation in locations of segment joints along the body, particularly in the anterior region. 4) Between different swimming behaviours, there are notable variations in segment numbers, along with variation in anterior joint locations. 5) In trout, swimming speed does not affect joint locations and segment numbers. The findings presented here provide a mechanistic understanding of relationships between kinematics and different fish attributes. These two methods and the models they generate could be applied in several open research areas, including construction of robot fish, investigation of fish muscle activity, and frameworks for lifeforms in virtual media.