LAUSR

Quantification of nocturnal migration of birds through moon watching is a technique ripe for modernization with superior computational power. In this paper, data collected by a motorized telescope mount was analyzed using both video observation by trained observers and modernized approach using computer vision. The more advanced data extraction used the OpenCV library of computer vision tools to identify bird silhouettes by means of image stabilization and background subtraction. The silhouettes were sanitized and analyzed in sequence to produce stacked relationships between temporally-close contours, discriminating birds from noise based on the assumption that birds migrate in stable paths. The flight ceiling of the birds was determined by extracting relevant correlation coefficient data from doppler radar co-located with the LunAero instrument in Norman, OK USA using a method with low-computational overhead. The bird paths and flight ceiling were combined with lunar ephemera to provide input for the original method used for nocturnal migration quantification as well as an enhanced version of the same method with more advanced computational tools. We found that the manual quantification of migration activity detected 16 300 birds/kmh heading northwest from 110○; whereas the automated analysis reported a density of 43 794 birds/kmh heading northwest from 106.67○. Hence, there was agreement with regard to flight direction, but the automated method overestimated migration density by ∼3 ×. The reasons for the discrepancy between flight path detection appeared to be due to a substantial amount of noise in the video data as well as a tendency for the computer vision analysis to split single flight paths into two or more segments. The authors discuss ongoing innovations aimed at addressing these methodological challenges.