Deep Computer Vision para la obtención de métricas comerciales en sucursales

Abstract

Customer behavior in physical branches constitutes a relevant metric for operational decision-making, such as staff allocation and the optimization of information point layouts. This project implements a person detection and tracking system based on Deep Learning-driven computer vision techniques, aimed at collecting commercial metrics from video recordings at UNIDROGAS S.A.S. branches. A connection was established between the cameras and a centralized server responsible for executing a detection model trained under the YOLO architecture, using real images from the branches to improve customer detection and the distinction between customers and employees. The generated information is stored on the server and used for the implementation of visualization tools. As a result, it was possible to consistently measure customer flow within the branches, as well as person counts in different areas of the establishment. This allowed for the generation of heatmaps that support the analysis of customers' spatial behavior and aid in operational decision-making.