Engdahl, Julia. Developing an automated analysis of fish migration video using computer vision algorithms. Retrieved from https://doi.org/doi:10.7282/t3-yavb-3m08
DescriptionDams, fishways, and other hydropower structures have been utilized for centuries and have largely been unevaluated regarding their effects on the surrounding fish populations. Conventional methods (seining and angling) used for such evaluation are often costly and time consuming. The aim of this project is to decrease analysis time by automating fish detection with the use of video monitoring and evaluating two computer vision algorithms: background subtraction and machine learning algorithm You Only Look Once (YOLO). We evaluated these algorithms on video data collected from the Island Farm Weir on the Raritan River in New Jersey. Our results indicate that background subtraction models need to be adaptive with respect to time due to the dynamic aquatic environment, and YOLO needs to be trained and tested on the user’s specific case study dataset for optimal results.