Many research areas rely on the ability to observe insect position and pose through markerless tracking. Neural networks provide more robust tracking systems than traditional computer vision-based systems. However, they require many human-hours to accurately label the training data. In this research, we explore the feasibility of using traditional computer vision techniques to label the training data automatically. This research suggests that, although a fully-automated labeling system may not be accurate enough to be used alone, such a system could be used as a first pass to create reasonably accurate labels, with a human performing a less work-intensive second pass to correct some errors. Such a hybrid approach would enable the labeling of more data with less human effort, given that the majority of the labeling burden would be shouldered by the automatic labeling system.