A computer-vision based automatic monitoring technique for detecting multi-individual movement in insect groups through planar motion data

Insect group behavior has been widely adopted for monitoring social interactions in insects recently. Based on deep learning and object tracking methods we developed a behavior monitoring system for automatically recording the movement trajectories of each individual in multi-individual groups of insects concurrently in the laboratory conditions. Physical descriptions including movement direction and speed of each individual within the group (maximum of 20 species) are continuously acquired by the developed computer vision algorithm after solving the problems of occlusion, aggregation, partial image, noise, etc. Moreover, the system could provide information on relative movement between/among individuals including the center of groups and velocity during the course of either forming or breaking-up the groups. The developed system was demonstrated with some indicator species including Drosophila and Thrips. The identified inter-individual movements would provide essential information on revealing sociality in insects.