Abstract
Flight patterns of bats have been tracked and studied before, but all tracking has been done manually. Throughout the semester, researchers aimed to automate the tracking of flight patterns in bats in order to more efficiently study behavior in both lab and field conditions. Researchers used an app DLTdv8 within MATLAB in order to analyze videos of bats flying. With DLTdv8 points on the bat such as the wingtips, body, feet, and thumb joints were able to be digitized frame by frame and then converted into datastores. The datastores were all digitized manually in order to give the computer the best possible data to learn. These datastores were then used to create networks through the Deep Learning capabilities of DLTdv8; the networks are a collection of datastores that the computer learns in order to apply to another video. A network can be a single datastore but for the sake of versatility in the networks, many networks had several datastores used to train it. Using DLTvd8 to track bat flight patterns was partially successful. The accuracies of the networks varied depending on how many datastores were used to train the network and also in which video the networks were applied. Some issues did occur when trying to apply networks to lab conditions and field conditions. In lab, learning that the networks needed to be tested on videos that were not used for training in order to see how it truly preforms was an important realization. In field the main issue that had to be overcome was isolating the bat in the videos. The field videos are not as clear and have more background objects that the network tried to misidentify as points on the bat so a cropping tool was used to fix this issue. This research is not done and the exploration of DLTdv8 is still ongoing; the exploration that was accomplished through this research shows some of the ways that DLTdv8 can be used to help automate tracking the of bats and the errors that arose while trying to accomplish that goal. Ideally this program would be able to track objects through space based on trained data, but more realistically in the near future it can help minimize the amount of automated tracking that is required to study the flight patterns and behavior of bats.