METHOD OF ROBUST MAIN BODY PARTS ESTIMATION USING DESCRIPTOR AND MACHINE LEARNING FOR POSE RECOGNITION

摘要

PURPOSE: A major body part estimating method which uses a directional technician and a machine learning for a posture recognition is provided to accurately estimate a major body part even in case the major body part is overlapped with other body parts by obtaining an outer edge and an inner edge and estimating the major body part by using a human image which has depth information. CONSTITUTION: A program of a computer device receives a human image (S1000). The program detects the edge of the human image (S2000). The program detects end points of a major body part from a generated edge image (S3000). The program detects predicted points of the major body part (S4000). The program obtains the central point of the detected predicted points as a candidate point of the major body part (S5000). The program obtains the edge distance information of the candidate point (S6000). The program estimates the body part of the candidate point by a comparison of the candidate point and learning data (S7000). [Reference numerals] (AA) Start; (BB) End; (S100) Generate learning data by using feature of head, hand and foot; (S1000) Input human image; (S2000) Detect image edge; (S3000) Detect end points of a major body part; (S4000) Detect predicted points of the major body part; (S5000) Obtain candidate point of the major body part; (S6000) Obtain the edge distance information of the candidate point; (S7000) Estimate body part of the candidate point by a comparison of the candidate point and learning data