CLASSIFICATION DEVICE, CLASSIFICATION METHOD, ATTRIBUTE RECOGNITION DEVICE, AND MACHINE LEARNING DEVICE

• Main Author: NAKAO NOZOMI
• Format: Patent
• Language: English; Japanese
• Published: 04.07.2019
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

To classify a subject to be recognized into a plurality of attribute classes with high accuracy.SOLUTION: A classification device for classifying a subject into a plurality of attribute classes related to predetermined attributes using a captured image captured by a camera installed above the subject to be recognized, includes: a storage unit configured to store divided area information representing a plurality of divided areas generated by dividing the imaging range of the camera according to a distance from a position immediately below the camera; an image processing unit configured to extract a subject area image that is an image of an area including the subject from the captured image; and a classifier configured to output each probability that the subject belongs to the plurality of attribute classes for each of the plurality of divided areas using the subject area image.SELECTED DRAWING: Figure 1 【課題】認識対象の被写体を複数の属性クラスに精度良く分類する。【解決手段】認識対象の被写体より上方に設置されたカメラで撮影された撮影画像を用いて、被写体を所定の属性に関する複数の属性クラスに分類する分類装置であって、カメラの撮影範囲をカメラの直下からの距離に応じて分割して生成された複数の分割領域を表す分割領域情報を記憶する記憶部と、撮影画像から被写体を含む領域の画像である被写体領域画像を抽出する画像処理部と、被写体領域画像を用いて、複数の分割領域ごとに、被写体が複数の属性クラスに属する各確率を出力する分類器と、を備えるものである。【選択図】図１