Кордюков Кирилл Сергеевич
Deep Learning of Energy and Nutritional Value in Food Analysis from Video
Науки о данных
The work is devoted to the search and implementation of a method for assessing the calorie content of food by video. The main difficulty in finding such a method lies in the assumption that the implemented approach should be able to determine the caloric value of food from one image and work in real time. To solve this problem, an algorithm was developed that can be divided into 4 stages. The first step is to find the location of the food in the image (localization), as well as determine the name of the product (detection). The second stage consists in highlighting the exact area of the product in the image obtained after the first stage (segmentation). At the third stage, the volume of food allocated at the second stage is estimated by assessing the depth of the image. And at the last stage, the caloric value of food is determined directly from the received volume.
Текст работы (работа добавлена 29 октября 2019г.)