Binocular stereoscopic vision type food identification method

Abstract

The invention discloses a binocular stereoscopic vision type food identification method. The binocular stereoscopic vision type food identification method comprises the following steps: S1, acquiringan image 1 and an image 2 of food images by utilizing double cameras; S2, selecting K reference points and reading pixel coordinates; calibrating a distance relation between the reference points and the cameras; S3, calculating inner and outer parameters of the cameras and carrying out binocular correction on the images which are acquired through the double cameras according to the inner and outerparameters; S4, carrying out stereoscopic matching on the images and calculating a disparity map; calculating a depth map according to the disparity map; S5, carrying out characteristic extraction according to the depth map; obtaining an identification result after carrying out the characteristic extraction. According to the binocular stereoscopic vision type food identification method disclosedby the invention, the food image is acquired at different angles through setting the double cameras and a stereoscopic space is established on the basis of two angle information; overlapped and shielded foods are recognized in the stereoscopic space, so that the problems in a food identification process of a single camera at present are solved.

Description

technical field [0001] The invention relates to a food recognition method, in particular to a binocular stereo vision food recognition method. Background technique [0002] With the development of intelligent technology, based on the deep neural network, the method of using convolution calculation to extract the features in the input food image, and classifying the food according to the extracted features has been gradually applied. The food recognition method based on the deep neural network usually uses the camera image to extract the food image features according to the convolution calculation to classify the food. However, the existing food recognition methods all use a single camera, the extracted food image is input into the neural network model, and the recognition is performed according to the features extracted by the convolution operation. This image obtained by a single camera can only contain a certain food. For the food recognition scene with overlapping or occ...

AI Technical Summary

Problems solved by technology

[0004] The present invention provides a binocular stereo vision food recognition method, aiming to solve the problem that the current food recognition method cannot be accurately recognized when the food is blocked, and is not suitable for recognition in complex environments such as canteens

Images