A Computer Vision Algorithm for Omnidirectional Bee Counting at Langstroth Beehive Entrances

摘要

A computer vision algorithm is proposed for omnidirectional bee counting in images of Langstroth beehive entrances captured hi situ with a miniature camera connected to a multi-sensor, solar-powered electronic beehive monitoring device. The algorithm consists of three stages: pre-processing, landing pad identification, and omnidirectional bee counting. In the pre-processing stage, an approximate image region where the landing pad is likely to be is cropped and the brightness of the cropped image adjusted. The landing pad identification is obtained through iterative reduction of the cropped image to the actual landing pad. Omnidirectional bee counts are computed by dividing the total number of bee pixels by the average number of pixels occupied by individual bees. The algorithm was evaluated on 1,781 images from two electronic beehive monitoring devices deployed hi Langstroth beehives with live honeybees and achieved an accuracy of over 80 per cent compared to the ground truth obtained from human evaluators.