Movement tracking is a commonly used method to measure locomotion and environmental interaction. It is a highly versatile tool allowing for measurements involving memory and cognition, social interaction, pharmacological effects, anxiety and depression, and much more. Unfortunately, the algorithms commonly used for animal tracking struggle with increased environmental complexity such as uneven or dim lighting, interference from additional recording hardware (i.e., optogenetic or neurophysiological manipulation), and experimenter interaction. Thus, Guanglong Sun and colleagues have developed DeepBhvTracking, a novel tracking algorithm that combines deep-learning and background subtraction.
Using the convolutional neural network-based You Only Look Once (YOLO) algorithm in combination with standard background subtraction, DeepBhvTracking has a faster training time and superior accuracy and much faster tracking speed than other deep-learning based tracking software. DeepBhvTracking can also track multiple color labeled animals at once allowing for measurement of social interaction. Therefore, DeepBhvTracking is a tracking software that is fast, versatile, and accurate.