Image-Based Estimation of Real Food Size for Accurate Food Calorie Estimation

摘要

In this paper, we review our works on image-based estimation of real size of foods for accurate food calorie estimation which including three existing works and two new works: (1) "CalorieCam" which is a system to estimate real food size based on a reference object, (2) Region segmentation based food calorie estimation, (3) "AR DeepCalo-rieCam V2" which is based on inertial visual odometry built in the iOS ARKit library, (4) "DepthCalorieCam" which employs stereo cameras on iPhone X/XS, and (5) "RiceCalorieCam" which exploits rice grains as reference objects. Especially, the last two new methods achieved 10% or less estimation error, which was enough for robust food calorie estimation.