Real Time Retail Analytics with Computer Vision

The retail industry is in a transformation as a result of shifting customer expectations and the advancement of technology via the internet. While more and more customers prefer to buy online, it is critical for physical stores to meet customers' expectations in order to maintain profitability. Therefore, customer data obtained from retail stores will reveal the dealer's performance and areas for improvement. Data on consumer traffic, customer pathways, and demographic factors will be acquired using the developed computer vision system. These data may be analyzed and used in store design and shelf alignment, demand research and stock optimization, consumer profiles, marketing and campaigns, and can provide information about key performance indicators. In this study, ResNet-18 based algorithms were built to acquire data on the number of customers, customer entry-exit rate, age distribution, and customer gender using image processing and deep learning methods through the use of the camera and embedded system placed in the dealers. The algorithms have been pruned to get a high performance of 13 frames per second on an NVIDIA Jetson Nano device with edge computing. Main goal is to analyze the data provided to the cloud and convey it to the sales and marketing departments, therefore improving retail shop performance.