Honeybee In-Out Monitoring System by Object Recognition and Tracking from Real-Time Webcams

A new honeybee in-out monitoring system is proposed using real-time deep-learning based image recognition and tracking. The specific design of beehive gate is turned out to be an important factor for accurate bee movement monitoring. We check a series of beehive gate designs for the monitoring system. A novel gate design employing heart valve structure is proposed for ensuring one-way traffic for the bees as well as one-at-a-time gate passing, resulting in an improved bee detection accuracy. As for the deep-learning based image recognition framework, YOLOv4 is used in the proposed system for a better honeybee-detection accuracy as well as a faster detection in comparison to YOLOv3 which was employed for our previous study. In addition, DeepSORT algorithm is employed for a reliable tracking of the detected honeybees. In our experiments the proposed honeybee monitoring system exhibited 99.5% detection accuracy, while our previous system resulted in 97.5% in the same settings.