BEHAVIOR RECOGNITION LEARNING DEVICE, BEHAVIOR RECOGNITION LEARNING METHOD, BEHAVIOR RECOGNITION DEVICE, AND PROGRAM

摘要

The present invention enables learning to obtain a behavior recognizer capable of performing behavior recognition at high accuracy by using a small amount of learning data.　An input unit 101 receives inputs of a learning video and behavior labels indicating behaviors of an object. A detection unit 102 detects, for each of frame images included in the learning video, a plurality of objects included in the frame image. A direction calculation unit 103 calculates the direction of a reference object that is an object defined as a reference among the detect plurality of objects. A normalization unit 104 normalizes the learning video such that positional relationships between the reference object and the other objects become predetermined relationships. An optimization unit 106 optimizes parameters of the behavior recognizer, on the basis of inferred behaviors obtained by input of the learning video normalized by the behavior recognizer for inferring behaviors of the objects in the inputted video and the behaviors indicated by the behavior labels.