Using Computer Vision And Volunteer Computing To Analyze Avian Nesting Patterns And Reduce Scientist Workload

This paper examines the use of feature detection and background subtraction algorithms to classify and detect events of interest within uncontrolled outdoor avian nesting video from the Wildlife@Home project. We tested feature detection using Speeded Up Robust Features (SURF) and a Support Vector Machine (SVM) along with four background subtraction algorithms — Mixture of Guassians (MOG), Running Gaussian Average (AccAvg), ViBe, and Pixel-Based Adaptive Segmentation (PBAS) — as methods to automatically detect and classify events from surveillance cameras. AccAvg and modified PBAS are shown to provide robust results and compensate for issues caused by cryptic coloration of the monitored species. Both methods utilize the Berkeley Open Infrastructure for Network Computing (BOINC) in order to provide the resources to be able to analyze the 68,000+ hours of video in the Wildlife@Home project in a reasonable amount of time. The feature detection technique failed to handle the many challenges found in the low quality uncontrolled outdoor video. The background subtraction work with AccAvg and the modified version of PBAS is shown to provide more accurate detection of events