Rapid technological advancements in the retail industry have enabled organizations to leverage innovative solutions for improving customer experience and employee productivity in multiple retail processes. One such area of advancement is in-store check-out experience. Typically, customers have to wait a long time in queues to get their purchases processed and make payments. According to research by TechRepublic and ZDnet, slow lines cost approximately $37.7 billion in potential sales per year.
Our client, a global retailer, started an initiative to leverage these technology advancements and implement a check-out automation solution. Resolve Tech team worked with the client’s team in developing the solution and implementing it in a pilot store. The pilot store implementation was to demonstrate the solution that helps customers to simply pick items they need and walk out. Using an app, customers can gain entry to the store. The credit card linked to the app would be charged for the purchase automatically and a receipt of purchased items would be sent to them electronically. RTS team was involved in activities such as Computer Vision algorithms development, Neural network training, Model evaluation, Selection & training, GPU builds & image pre-processing, and Accuracy evaluation.
Overhead cameras, weight sensors and deep learning technology to detect merchandise that shoppers take from or return to shelves and keep track of the items selected in a virtual cart.
Design, develop, and implement novel computer vision algorithms for realizing unique use cases using deep learning frameworks such as TensorFlow, Keras, PyTorch, Caffe etc.
Experience end to end self-service with a custom mobile app linked with credit card info for electronic payment and receipts
Pre-process training image dataset using tools such as OpenCV, Skimage etc. to increase accuracy and reduce time to learn for the models
Use object contours and backgrounds to generate augmented data in Python for simultaneous & accurate image recognition
Solve problems like human pose estimation, object detection, and face recognition by training neural nets