This thesis introduces the Remote High-definition Visual Monitoring (RHVM) system. This system provides an affordable and sophisticated alternative to current methods of visual monitoring of cetaceans at sea.There are several scenarios that require monitoring of marine mammals at sea; one of which includes efforts made to mitigate the effect of man-made noise (e.g. during seismic surveys) on the animals. Often used visual methods relies solely on humans (experts called Marine Mammal Observers) to detect the animals and subsequently estimate distance to them. In addition to problems caused by poor visibility at night, fog and fatigue, estimating distance at sea with the naked eyes is very difficult and is often guess work. Unsophisticated distance estimation methods, such as sighting stick, are not very accurate or precise and can result in unnecessary and expensive delays to the surveys or endanger animals.These problems are addressed by combining the application of robust computer vision algorithms with relative cheap off the shelf sensors and the latest telecommunication system. The sea environment presents very peculiar challenges to computer vision methods due to constantly changing atmospheric conditions, unpredictable movement of the vessel and very cluttered image scenes due to the waves. A highly flexible multi-sensors system has been designed and tested.A Real-time Automated Distance Estimation at Sea (RADES) algorithm has been developed for objective and recordable distance estimation at sea. The system tracks the vessel global orientation by detecting the horizon in images from a camera. Although horizon detection techniques have been studied in the past, the problem of real-time detection and ability to cope with a wide variety of conditions has not been effectively dealt with. The horizon detection technique developed here is made robust to weather effects by the application of the dark channel prior for pre-processing and robust to temporary occlusion by fusing visual measurement with inertial sensors data. Detailed mathematical analysis of the distance estimation technique is also given and a formula for estimating the resolution of the system is presented for the first time.A new algorithm for Automated Recognition of Cetaceans at Sea (ARCS) has also been developed. The algorithm relies on a novel application of morphological operation that adapt to the content of the image scene for extraction of whales blows. The algorithm is capable of coping with a considerable amount of noise in the challenging sea environments; using Support Vector Machines (SVM) for classification. Steps towards training the SVM includes an effective data cleaning step based on Tomek Links and a scheme for dealing with a highly imbalanced data set is given.
|Date of Award||1 Aug 2017|
|Supervisor||Adrian Evans (Supervisor) & Robert Watson (Supervisor)|