Our objective was to build a connected intelligent shopping cart/basket, which will detect, which products have been placed in it and will generate shopping recommendations for the current cart user. We have used the state-of-the-art, real-time object detection system - YOLOv2, which deep architecture has been reduced to accelerate the evaluation on the Raspberry PI device and the Intel® Movidius.
During the talk we will show our project live and we will also talk more about innovative technical aspects such as the Intel® Movidius™ Neural Compute Stick, AWS IoT and AWS Lambda.
Our solution has been also presented at the O'Reilly Artificial Intelligence Conference 2018 in London.
We have used several innovations in our project:
- Intel® Movidius™ Neural Compute Stick
- AWS IoT Core - AWS Lambda
- Docker
- Tensorflow
- OpenCV
- Intel DevCloud
- Raspberry Pi
- AWSIoTPythonSDK library
Our algorithm recognised the product even when only the back of the package was visible. It means that the logo doesn't have to be visible. Only a texture and a shape will be enough. We`ve observed nearly 100% accuracy. It was enough to put the product in the basket, without paying attention to its arrangement in space.
Traditional (off-line) retailers in the self-service stores do not have many opportunities to offer additional products (basicaly only by the cashier). Our solution gives the off-line retailer additional opportunieties to advertise their products in the very simmilar fashion, like on-line sellers (i.e. based on the products already placed in a shopping basket/cart).
