3D Reconstruction with SFM-MVS Method for Food Volume Estimation

Abstract

The growing need for accurately understanding calorie and nutrient intake highlights the limitations of traditional methods for assessing food portions. Advanced computer vision technologies, mainly 3D reconstruction methods, provide a more precise and automated approach to estimating food volume. This study focuses on developing a cutting-edge system that creates Three-Dimensional (3D) food images using the Structure From Motion-Multi View Stereo (SFM-MVS) method. Detailed volume estimation is performed after constructing the 3D model to ensure accurate measurements. Using a sophisticated mobile application to capture images from multiple angles, the system undertakes a comprehensive 3D reconstruction of food items. This complex process is enhanced by subsequent slicing and segmentation, allowing for detailed extraction and precise volume calculation of each food component. The system has undergone rigorous testing on various food types, consistently showing a volume estimation error rate below 10%, thus significantly improving the accuracy of food volume estimation. This research significantly advances automatic diet monitoring and calorie consumption management. By leveraging state-of-the-art 3D reconstruction techniques, the system effectively overcomes the limitations of traditional methods, providing a reliable, efficient, and user-friendly approach to dietary assessment. Consequently, it supports more effective nutritional planning and health management, meeting the growing demand for precise and automated dietary monitoring solutions. The impact of this research is extensive, offering considerable benefits for health professionals, nutritionists, and individuals seeking accurate nutritional information to support better health outcomes.