FOODCAM: A Novel Structured Light-Stereo Imaging System for Food Portion Size Estimation

Area of Science:

• Biomedical Engineering
• Computer Vision
• Nutritional Science

Background:

• Traditional food portion size estimation (FPSE) methods lack accuracy.
• Existing imaging-based FPSE often requires dimensional cues or manual input, limiting their practicality.

Purpose of the Study:

• To introduce the FOODCAM (Food Intake Monitoring Camera), a novel device for accurate, reference-free food portion size estimation.
• To present a methodology for 3-D food reconstruction and volume estimation using the FOODCAM system.

Main Methods:

• Developed a passive, multispectral, motion-activated stereo camera system (FOODCAM) with integrated structured light.
• Implemented a volume estimation methodology based on 3-D reconstruction from stereo image pairs.
• Validated the system using five complex food models: banana, brownie, chickpeas, French fries, and popcorn.

Main Results:

• The FOODCAM system achieved an average food portion size estimation accuracy of 94.4%.
• The methodology demonstrated effective 3-D reconstruction and volume calculation for diverse food shapes.
• The system operates effectively without requiring dimensional references in a fixed setup.

Conclusions:

• The FOODCAM system provides a highly accurate and practical solution for food portion size estimation.
• This technology has significant potential for application in diet and eating behavior research.
• The reference-free methodology enhances the usability of imaging-based FPSE in real-world scenarios.