Analyzing Bee Behavior Through Video Tracking Using Computer Vision Techniques

Bees are responsible for a large part of the pollination of plant species and act as bioindicators in evaluating phytosanitary products. Research on bee behavior and mortality can be carried out in laboratories with a controlled environment. Still, even so, behavioral assessment, when performed visually by an expert, becomes time-consuming and susceptible to failures, which justifies the creation of automated systems for this purpose. This article presents a tool based on Computer Vision to support this activity. The YOLO/Darknet architecture was used to detect bees in videos, and experiments were carried out with eight pre-trained models, obtaining adequate results of insect detection, which were used to develop an interface that provides graphs and reports: heat maps, walking maps, the percentage of the area explored, the time spent moving, and the speed at which individuals move. Experts can interpret the information produced by the proposed tool to understand bee behavior.