AI-based Customer Behavior Analytics System using Edge Computing Device

Abstract

Customer behavior can be analyzed using an artificial intelligence algorithm, specifically object detection algorithm. However, the implementations are usually either using a cloud computing-based system or a high-end PC. These existing solutions require a high-traffic internet connection or a large-sized GPU. In this paper, we propose a system implementation using an edge-computation approach. The purpose of this paper is to port an existing deep learning-based in-store traffic monitoring system that is used for evaluating retail performance into an embedded computer, specifically NVIDIA Jetson Nano. This is done in order to minimize the cost, power consumption, and space needed to implement this system on a larger scale. The system can run in 10 frames per second using Single Shot Detector MobileNet v2 architecture.