You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Bab-Al
AI Nutritrack : Harnessing Vision for Personalized Nutrition
🥗 Nutritional Analysis: Identify ingredients from food photos.
🍳 Recipe Recommendations: Personalized suggestions based on user preferences.
đź“Š Health Tracking: Graphs for better dietary management.
🤝 Raspberry Pi Integration: Meal logging for elderly and caregivers.
🤒 Motivation
Digital Culture Impact
Nutritional Imbalance
Healthy Eating Focus
Caregiver Support
🎯 Target
1. 🍎 Healthy Eating Habits : Promotes balanced diets and supports individuals aiming for long-term health improvements 2. 🧓 Patients or Elderly : Assists physically challenged individuals and helps caregivers effectively manage dietary needs
🌌 Flow Chart
Techniques
🔧 Tools
Category
Technologies
AI
Frontend
Backend
Embedded System
Others
🛠️ Architecture
🧶 Algorithm
1. Raspberry Pi Kernel Module for Camera
When the user presses a button, the connected Raspberry Pi takes a picture using the camera module.
The captured image is automatically sent to our AI server for processing.
The detected food name and nutritional information are then sent to our backend server via a POST request.
Consequently, the user's family or caregivers can view the meal records through our iOS app.
2. Food Object Detection & Nutrient Estimation
The received image file is resized to 640x640px, and the filename is modified to include the current time.
The image is saved on Google Drive in a specified folder.
Food object detection is performed using a trained model, and the results are stored in a `.txt` file.
The system reads the result and extracts the class codes of the detected food items.
The same image is then used for quantity estimation using a different model, which estimates the number of food items based on bounding boxes and reference objects.
The quantity and nutritional content of the detected food are calculated and sent as a JSON response to the client.
3. Recipe Recommendation
The recommendation system provides two recipes based on the user's information, such as age, gender, and dietary preferences.
First, food data is filtered based on a 7:3 ratio between ingredients and tags.
Nutritional components like carbohydrates, proteins, and fats are converted into vectors, and cosine similarity with the user's vector is used to generate the first recommendation.
The second recommendation is based on the first recommendation and other users' consumption patterns, using a NGCF-based model in the RecBole framework to predict similar items.
