Food Volume Estimation Using 3D Shape Approximation for Medication Management Support

The authors have been developing a food volume estimation technique in order to remind medication time. In the previous study, they used a depth image and recognized the decrease of food based on the change of distance to food. In this study, the authors attempted to use a 3D point cloud and directly estimate the volume of food. Firstly, points not related to food were deleted by preprocessing. Secondly, if there are some objects on a dish, they were separated by using a clustering method. The surface area viewed from the camera and the height from the dish surface were calculated for each object. Finally, the volume of each object was calculated by multiplying the area and the height. An experiment showed that food volume can be estimated with high accuracy by removing outliers.