A Low-cost Raspberry Pi and Deep Learning System for Customer Analysis

Knowing what customer’s interest when they visit a store may help them increasing its efficiency and turnover. Using current technology advancements, this project intends to create a system that recognizes the presence of consumers visiting a store by constructing designated bounding boxes to compute how long the client stays in each retail area. Our system was developed utilizing the COCO dataset using SSD MobileNet V2 system based on Raspberry Pi. The Raspberry Pi is linked to a camera, which records customer activity in each indicated location and calculates the customer's time spent. A low-cost specification allows the system to be easily implemented in the real environment. From the simulations with two bounding boxes, in region 1 there are 7 people with an average time spent of 12 seconds and in region 2 there are 13 people with an average time spent reaching 20 seconds with an accuracy of 87.5 percent and 70 percent, respectively.