THREE-DIMENSIONAL PERSON ATTITUDE ESTIMATION APPARATUS, METHOD, AND PROGRAM

• Main Authors: TASAKA KAZUYUKI, NAKATSUKA TOMOHIRO
• Format: Patent
• Language: English; Japanese
• Published: 22.06.2022
• Subjects: CALCULATING; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

To provide a three-dimensional person attitude estimation apparatus, method and program for accurately estimating an attitude of a person including the amount of twist rotation of four limbs with less computational complexity.SOLUTION: In a three-dimensional person attitude estimation apparatus 1, a joint point three-dimensional coordinate estimation unit 20 estimates three-dimensional coordinates for each of joint points on the basis of an image of a subject person. A twist rotation amount estimation unit 30 estimates the amount of twist rotation of four limbs for each of the joint points on the basis of the image of the subject person. A body model storage unit 40 stores representative joint points of a person with definitions of four limbs connecting the joint points and a restraint condition, as a body model. An estimation unit 50 estimates an attitude parameter of the subject person by modifying the body model so as to minimize a difference between the three-dimensional coordinates of the joint points and the twist rotation amount of the four limbs of the body model and results of estimating the three-dimensional coordinates and the twist rotation amount.SELECTED DRAWING: Figure 1 【課題】人物の四肢のひねり回転量も含めた姿勢を少ない計算量で正しく推定できる三次元人物姿勢推定装置、方法およびプログラムを提供する。【解決手段】三次元人物姿勢推定装置1において、関節点三次元座標推定部20は、被写体人物の画像に基づいて関節点ごとに三次元座標を推定する。ひねり回転量推定部30は、被写体人物の画像に基づいて関節点ごとに四肢のひねり回転量を推定する。身体モデル記憶部40には、人物の代表的な関節点が各関節点を連結する四肢（リブ）の定義や拘束条件と共に身体モデルとして記憶されている。姿勢パラメータを推定部50は、身体モデルの各関節点の三次元座標および四肢のひねり回転量と前記三次元座標およびひねり回転量の各推定結果との差分が最小化するように身体モデルを変形させて被写体人物の姿勢パラメータを推定する。【選択図】図1