Abstract:
In this work we propose a simple yet innovative algorithm for beverage detection (brand and bottles/cans amount) through the visual analysis of a cooler shelf picture snapped by a wideangle cheap camera. This algorithm relies on a loose Chamfer matching, for a first check based on shape detection, and a stricter color matching based on simple 3D color histograms, for the last confirm. The algorithm is optimized by using 3D modeling techniques for precise template generation and a space management system which allows faster image scan and prevents possible detections interpenetration. Further accuracy is achieved by splitting a beverage into its main characterizing parts, processing them independently and considering the results as a whole. Occlusion is dealt by building an occlusion mask which keeps track of the image portions occupied by the detections and by masking with this mask the templates occluded parts. To achieve a better background invariance, it is enforced a minimal distance between the results of each detection and the results obtained by the application of the same template onto a reference background image. Finally the results are fixed through the application of ad-hoc heuristics based on non visual information and common people behavior. The algorithm role is to support VeBox, a specialized IoT telemetry tool for sales infrastructures, by collecting sales data from coolers in order to provide material for the sale management functions.
