Automatic classification of flying bird species using computer vision techniques

Bird species are recognised as important biodiversity indicators: they are responsive tochanges in sensitive ecosystems, whilst populations-level changes in behaviour are bothvisible and quantifiable. They are monitored by ecologists to determine factors causingpopulation fluctuation and to help conserve and manage threatened and endangeredspecies. Every five years, the health of bird population found in the UK are reviewedbased on data collected from various surveys.Currently, techniques used in surveying species include manual counting, Bioacousticsand computer vision. The latter is still under development by researchers. Hitherto,no computer vision technique has fully been deployed in the field for counting speciesas these techniques use high-quality and detailed images of stationary birds, which makethem impractical for deployment in the field, as most species in the field are in-flight andsometimes distant from the cameras field of view. Techniques such as manual and bioacousticsare the most frequently used but they can also become impractical, particularlywhen counting densely populated migratory species. Manual techniques are labour intensivewhilst bioacoustics may be unusable when deployed for species that emit little or nosound.There is the need for automated systems for identifying species using computervision and machine learning techniques, specifically for surveying densely populatedmigratory species. However, currently, most systems are not fully automated and useonly appearance-based features for identification of species. Moreover, in the field,appearance-based features like colour may fade at a distance whilst motion-based featureswill remain discernible. Thus to achieve full automation, existing systems will haveto combine both appearance and motion features. The aim of this thesis is to contribute tothis problem by developing computer vision techniques which combine appearance andmotion features to robustly classify species, whilst in flight. It is believed that once this isachieved, with additional development, it will be able to support the surveying of speciesand their behaviour studies.The first focus of this research was to refine appearance features previously used inother related works for use in automatic classification of species in flight. The bird appearanceswere described using a group of seven proposed appearance features, whichhave not previously been used for bird species classification. The proposed features improvedthe classification rate when compared to state-of-the-art systems that were basedon appearance features alone (colour features).The second step was to extract motion features from videos of birds in flight, whichwere used for automatic classification. The motion of birds was described using a groupof six features, which have not previously been used for bird species classification. Theproposed motion features, when combined with the appearance features improved classificationrates compared with only appearance or motion features.The classification rates were further improved using feature selection techniques.There was an increase of between 2-6% of correct classification rates across all classifiers,which may be attributable directly to the use of motion features. The only motion featuresselected are the wing beat frequency and vicinity features irrespective of the method used.This shows how important these groups of features were to species classification. Furtheranalysis also revealed specific improvements in identifying species with similar visualappearance and that using the optimal motion features improve classification accuracysignificantly.We attempt a further improvement in classification accuracy, using majority voting.This was used to aggregate classification results across a set of video sub-sequences,which improved classification rates considerably. The results using the combined featureswith majority voting outperform those without majority voting by 3% and 6% on the sevenspecies and thirteen classes dataset respectively.Finally, a video dataset against which future work can be benchmarked has beencollated. This data set enables the evaluation of work against a set of 13 species, enablingeffective evaluation of automated species identification to date and a benchmarkfor further work in this area of research. The key contribution of this research is that aspecies classification system was developed, which combines motion and appearance featuresand evaluated it against existing appearance-only-based methods. This is not onlythe first work to combine features in this way but also the first to apply a voting techniqueto improve classification performance across an entire video sequence.