CPC G16H 20/60 (2018.01) [G06F 18/24323 (2023.01); G06N 3/08 (2013.01); G06V 10/82 (2022.01); G06V 20/46 (2022.01); G06V 20/68 (2022.01); G06V 30/19173 (2022.01); G06V 30/2247 (2022.01); G06V 30/412 (2022.01)] | 26 Claims |
1. A food-recognition method, comprising:
inputting each frame, of a plurality of frames of a video stream, into a multiple-object detector to obtain (i) a bounding box identifying where each food item, of one or more food items, appears within said each frame, and (ii) a predicted multiple-object class identifying said each food item with a multiple-object probability;
for each bounding box:
cropping said each frame into a cropped frame based on said each bounding box;
inputting the cropped frame into a classifier to obtain one or more predicted classifier classes that identify said each food item with corresponding one or more classifier probabilities; and
determining a food identity of said each food item based on the predicted multiple-object class, the multiple-object probability, the one or more predicted classifier classes, and the corresponding one or more classifier probabilities;
creating a frame list of the one or more food identities determined from said each frame;
enqueuing at least one of the one or more food identities in the frame list into a most-viewed queue of elements to create an updated most-viewed queue; and
outputting at least one of the one or more food identities in the updated most-viewed queue.
|