A no-cost retail video analytics data collector for data-driven practices enabilng on retail stores.
- 0 Collaborators
The retail-video-analytics-data-collector is a Python script that captures frames from your webcam, identifies any faces on it, runs each faces through several neural networks optimized with OpenVINO, gathers the age, gender, emotion and its head pose and stores it in a csv. ...learn more
Project status: Published/In Market
Intel Technologies
Intel NUC,
Intel Python,
Movidius NCS,
OpenVINO
Overview / Usage
This project solves the problem of capturing clients data while in a store. By, on each frame, capturing the age, head pose, emotion and gender of each person (and storing it in a .csv file), it lets the owner of the stores to get insights and data-driven decisions in real time- what will surely be very helpful in maximizing the store and employees performance.
An example of what can be done is using the gender, age and emotion to control the content exhibition in the store: by knowing the age, gender and emotion, the store owner can adapt the ads shown and music played in the sotre, for example. Also, it gives a preprocessed, scaled, one-hot encoded and normalized gathered data so it can be used to fit some machine learning model by labeling it with any kind of data.
Methodology / Approach
The approach is very simple: we run each frame in OpenCV face detector, and gather the coordinates of each face. Then sub-setting the tensor representation of the image, we get each faces image, which is preprocessed and run into several neural networks, so we can get the person's emotion, gender, age and head pose. We store it in a tuple and then into a Pandas DataFrame, sp we are able to save it easily in a csv. We also save a copy of the csv with preprocessed, scaled, normalized and one-hot encoded data, so it can be used for analytics and insights producing. The whole project was developed in a Intel NUC and tested with a Movidius NCS-1.
The whole project is powered with Intel Technology:
-For image capturing and frame-preprocessing, Intel OpenVINO accelerated OpenCV lib for Python is used. The same lib is also used for, in the real time video shown in screen, plot the inferences on each face.
-The neural nets used to infere from the image are OpenVINO Optimized Models, so it runs with a great performance even in edge devices (also it can be plugged in Intel NCS using command line arguments).
Technologies Used
-Intel Distribution for Python
-Intel Distribution for OpenVINO Toolkit
-OpenVINO toolit Model Optimizer
-Intel Movidius NCS
-Intel NUC
Repository
https://github.com/piEsposito/retail-video-analytics-data-collector