The application is split into three containerized components: one responsible for store traffic monitoring through people detection, one for detecting a shopper's gaze and sentiment, and one for checking the shelf inventory.
❶ A trained neural network detects people within a designated area by using the mobilenet-ssd model for inferencing, and then stores and displays the current number of people in the frame and the historical total.
❷ An inferencing pipeline uses a face detection model to identify the presence of a shopper looking at a product. This data is passed to an emotion recognition model and head-pose detection model to determine engagement and sentiment.
❸ The captured analytics data is sent to an InfluxDB for storage and further analysis, with a data-rich dashboard for visualization using Grafana.