On Edge Crowd Traffic Counting System using Deep Learning on Jetson Nano for Smart Retail Environment

Abstract

Most supermarket don’t have a crowd traffic counting systems to track and counting customer entering their shop lot. In practise, these methods are useful for businesses in managing customer flow and optimize staffing and marketing efforts. Furthermore, the information can be used to estimate the popularity of the shop with relation to people entering the shop lot. The information also useful for the shop owner in determining the renting value. Therefore, the paper presents a system for tracking and counting people in a retail store using on the edge device, a Jetson Nano board. The comparison of the performance of two algorithms for people detecting of YOLOv5 and MobileNet-SSD are used in this work, YOLOv5 is a state-of-the-art object detection model that is known for its accuracy, but it is computationally intensive and may not be suitable for running on resource-constrained devices such as the Jetson Nano. MobileNet-SSD is a lightweight object detection model that is designed to run efficiently on mobile devices and embedded systems. Next is to track people using SORT, SORT is a real-time multi-object tracking algorithm that is based on the Kalman filter and the Hungarian algorithm. The results show that YOLOv5 was able to achieve the highest accuracy in detecting and counting people, but it was slower than the other two algorithms. MobileNet-SSD was the fastest algorithm, but it had lower accuracy compared to YOLOv5. In conclusion, the choice of algorithm will depend on the trade-off between accuracy and computational resources, and SORT is a good option for real-time people counting on resource-constrained devices.