Human's Behavior Tracking in a Store Using Multiple Security Cameras

Recently, various Internet of Things (IoT) platforms offer cloud-based AI using their services like Machine Learning and Neural Networks that operate in their powerful computer engines and provide functionality to IoT devices. Artificial intelligence is widely used in the topics of computer vision for developing smart technologies and intelligent systems. The aim of this research is to create an automatic tool for tracking human's behavior in stores to identify the most popular places. Based on this information, more efficient product or commercial arrangements could be applied that are based on the latest tendencies. In this paper all humans as objects were localized by identifying human features using Gaussian mixture probability density model. After all images were collected and humans were labeled, the real-time human detection and tracking methods were realized. Several NN architectures were analyzed and compared. It appears that the pre-trained Custom Vision model reaches 80% accuracy in 1 hour of training at Microsoft Azure platform. The proposed technique allows not only to track people's behavior but also create a heatmap of the store providing the most visited places, where customers stop and pay more attention.