| Summary: | Recently, there has been an increase in biological research interest in fish, and zebrafish, as an efficient model in the investigation of a broad range of human diseases and genetic studies. Economically, the enormous number, low price and limited maintenance requirements of this fish species encouraged the researchers to use it extensively. The larva of this animal is also considered to be promising subjects for research that is not subject to same strict legal requirements as the adult fish. The importance of this animal in research has increased the demand for developing new computer vision tools and methods that could help researchers to perform more related investigations as well as understand behaviour for different experimental tests. Computer vision is an efficient, economical and non-intrusive tool that can be applied to research in aquatic laboratories and aquaculture environments. However, in marine applications, this technology is still facing big challenges due to the free-swimming nature and unpredictable behaviour of the fish. This thesis presents a suite of novel and cost-effective tools for fish tracking and behavioural analysis, sizing, and identification of individual zebrafish. These main contributions this work is outlined briefly as follows. The first part of this work deals with stimulation and physical activity analysis for fish larvae, a novel robust and automated multiple fish larva tracking system is proposed. The system is capable of tracking twenty-five fish larvae simultaneously and extracting all physical activity parameters such as; speed, acceleration, path, moved distance and active time. The system is used for further studies throughout local occurrence behaviour recognition and studying the behavioural of the fish larvae following electrical, chemical and thermal stimulation. The proposed tracking system has been adopted in the biologists' aquatic laboratory to be used as a robust tool for fish behaviour analysis when fish are exposed to several types of stimulation. In the second part of the work, two novel practical and cost-effective models; orthogonal and stereo systems are designed and implemented to estimate the length of small free-swimming fish using off-the-shelf-components. The designed models are accurate and easy to adapt use for small experimental tanks in laboratory settings. The models have been thoroughly tested and validated experimentally. The third part of this thesis offers novel non-contact methods for recognition of individual free-swimming fish. Such systems can significantly reduce experts efforts and time required for fish tagging process and also offer a real-time recognition technique that can be alternative to the existing tagging methods used in this field. Through the purposes of this suite of novel computer vision tools and models, this thesis has provided successful solutions for behavioural analysis, fish sizing, individual fish recognition related research problems. The proposed solutions addressed major research problems and provided novel and cost-effective solutions for these problems.
|
|---|
