A Robust Multiple Honeybee Tracking Method from Videos Captured at Beehive Entrance

Monitoring the health of honeybees is an important task of beekeepers to mitigate negative impacts that could happen to the beehives. In fact, the high density of honeybees at the beehive entrance and the complex movements lead to nonlinear motion and heavy occlusion challenges for honeybee detection and tracking. To tackle the aforementioned challenges, a method for honeybee detection and tracking that incorporates a real-time object detection module based on YOLOv5 and robust object tracking based on OC-SORT is proposed for honeybee detection and tracking from images captured at the beehive entrance. Extensive experiments have been conducted on a self-built dataset named VnBeeTracking. Experiment results confirm the outperformed results of the proposed method compared with other tracking-by-detection methods. The proposed method obtained 78.3% and 88.2% for two important metrics MOTA and MOTP in object tracking while maintaining a low rate of ID switch.

I. Introduction

Honeybees play a very important role in the ecosystem as well as the agricultural economy. According to [1], insects are responsible for about 80% of all pollination activities, of which bees account for more than 80% of total insect pollination activities and honeybees are the most important pollinators. The pollination of honeybees has contributed to maintaining the diversity of plant ecosystems and improving crop yields. Besides, there are many products obtained from honeybees such as honey, beeswax, pollen, etc. and all of them have high economic value. Therefore, in many parts of the world, honeybee farms have been built and are increasingly being expanded.