Method for calculating food calories based on binocular vision

Abstract

The present invention provides a method for calculating food calories based on binocular vision, falling within the technical field of computer vision and nutrition analysis. The method includes the steps of: firstly, using a calibrated binocular camera to capture food images and performing image preprocessing; and calculating an optimized disparity map by combining a semi-global block matching (SGBM) algorithm with weighted least squares (WLS) filtering, and introducing an edge-disparity collaborative network to improve the matching accuracy of foods with weak textures; subsequently, segmenting a food region via a U-Net network, outputting precise contours after edge optimization, and identifying specific categories of the foods using a lightweight classification network; furthermore, adaptively selecting an equivalent cylinder method or point cloud reconstruction method based on the food shape for three dimensional (3D) volume reconstruction, and converting the volume to mass by referencing a density database; and finally, querying a calorie database including cooking method information to obtain an accurate calorie value. In the present invention, the method is obviously superior to the related art, and provides an efficient and reliable solution for family diet management and health monitoring.

Images