Detection and Tracking of Livestock Herds from Aerial Video Sequences

Abstract

Autonomous herding research is becoming increasingly relevant. In this work, a model for sheep detection in herds from aerial video sequences is proposed, using the convolutional neural network Mask R-CNN. Several trainings with different datasets have been performed for achieving the model. An improvement in the detection metrics, through a visual tracking tool, allows not only detecting the individual sheeps in the herd, but also tracking them along the different frames in aerial video sequences. This system could be used, for example, in a drone to carry out livestock supervision, in addition to obtaining metrics that allow knowing the status of the herd. Finally, the method has been validated using several tests on images and videos of livestock in real outdoor environments.