| CPC G06V 40/23 (2022.01) [G06V 10/80 (2022.01)] | 3 Claims |

|
1. An action-model generation apparatus comprising:
processing circuitry
to perform learning with use of video data indicating a basic action of a movable body, as teacher data, analyze transition of a state of each of a plurality of parts of the movable body, and generate a basic-action model which is a model for recognizing the basic action and indicates the transition of the state of each part of the movable body;
to acquire the basic-action model generated, for each of a plurality of the basic actions; and
to perform learning with use of video data, as teacher data, indicating an advanced action which is an action more complex than the basic action, analyze the transition of the state of each of the plurality of parts of the movable object, generate as a provisional model, a model for recognizing the advanced action, which indicates the transition of the state of each part of the movable object, combine two or more basic-action models a result of combination of which resembles the provisional model, among a plurality of basic-action models, and generate an advanced-action model which is a model for recognizing the advanced action, wherein
the processing circuitry further
combines two or more basic-action models among the plurality of basic-action models, and generates a plurality of candidate models which are candidates for the advanced-action model, and
calculates similarity degrees between each of the plurality of candidate models and the provisional model, and designates as the advanced-action model, a candidate model whose similarity degree is the highest.
|