RU Hacks 2018
myCart
May 2018
Placed 2nd for developing an Android app that allows users to acquire body composition data and, through AI and real-time object detection, identifies grocery items to obtain personalized nutritional advice.
A user takes a picture of their grocery basket which will be analyzed by our trained object detection AI. Each item will be identified and a total list of nutritional value will be presented to the user. This total is evaluated to determine if it satisfies the user's personalized biometric requirements (based on user's height, weight, age, and sex).
The user is now better informed on whether the items their cart satisfy, fail to satisfy, or exceed their nutritional needs.
Android Studio:
• Compatible for API level 25+ to support majority of Android devices
• Built in camera access
• Saves images to app’s designated storage location on SD card
• Calculates nutritional parameters such as basal metabolic rate (BMR)
• Based on user specific data, provides recommendation on whether the items in the cart meet, exceed, or fail to satisfy daily nutritional needs
Microsoft Cognitive Toolkit (CNTK)
• Compatible with Keras Library
• Opted for CNTK over TensorFlow due to its better handling of streaming content, faster processing capability, and better performance when running on a single system
• Layers images using Faster R-CNN algorithm
• Achieved AI recognition to an average of 80% accuracy
