A real-time fine-grained visual monitoring system for retail store auditing

Abstract

Automated visual analysis of the object is of prime importance to realize the real-time concept of the internet of things. In this paper, we proposed a real-time fine grained visual analytics system for tracing the visibility of products on retail store shelves. The proposed visual monitoring system (VMS) is aimed to achieve high rates of product recognition, regardless of several real-time challenges like occlusion, different lightening conditions, product orientation etc. To address all these issues, the VMS collects the local feature descriptors which are scale invariant, rotational invariant and illumination invariant from training template images. Once, the testing image uploaded from any camera enabled device, the VMS extracts same local features and matches with the target feature descriptors for finegrained object recognition. This paper also covers the performance of various state-of-the-art local feature descriptors for object detection in context of retail store monitoring/tracking. The performance of the VMS is tested on real time retail shelve images. The results after investigation, the proposed fine-grained VMS shows approximately 90% accuracy in brand level detection.