Video-based Flight Pattern Mining of Birds in Controlled Tunnel Environments Using YOLOv8 and Data Analytics

Fine-scale flight behavior of birds has become an important area of neuro-flight studies, behavioral ecology and autonomous flight models. This paper describes a video-based data mining system to auto-detect bird trajectories, analyze the behavior of flight, and categorize flight behavior patterns. In 12 feet long, 4 feet wide tunnel, five budgie ( Melopsittacus undulatus) (Inki, Pinki, Ponki, Ronki, Donki) were filmed at high frame rates, cameras were placed six feet above the floor of the tunnel. Frames with birds were automatically identified with the help of the pre-trained YOLOv8 object detection model and frames with no birds were eliminated to minimize noise. The identified positions were transformed into real-life coordination and assigned the names of the tunnel zones (Front, Middle, Back) and the walls sides (Left, Right). A detailed dataset was developed to continue analysis. Flight behaviors were explored with the use of clustering, decision tree-based classification, trajectory scatter plots, and Sankey diagrams. Findings indicate clear spatial choices, including cluster-based sorting of flight behavior, and zone changes, which give informative explanations of individual and group behaviors. The novelty of the work is in the design of a fully automated reproducible pipeline of tunnel-based mining of bird flight behavior by combining detection, real-world mapping, and pattern analysis using data in a controlled laboratory setting.