overview of the datasets available specifically for posture analysis, emphasizing their appropriateness for training and assessing machine learning models in pose estimation tasks
1) The effective of CNN models, presenec of high-quality training data
The problem statements
1) Posture deviation
2) Measurment of postures' deviations concerns
3) Method to visulalize the results
Need of comprehensive analysis of pose deviation to integrate and synthesize the knowledge across multiple parameters.
OpenPose , Real time multi-person pose estimation (Carnegie Melon Univ.)
Bottom-up approch
1) 사진에 보이는 모든 관절 확인
2) 관절과 연결될수있는 모든 관계성을 확인
High accruacy 보장해서 유명.
Analyze 한 postural parameters
1) Neck inclination angle
-
C7 to ear tragus
2) Shoulder angle
- C7 to acromial angle (어깨봉우리)
3) Cranial rotation angle
- ear tragus to outer corner of eye
4) Neck-length / Shoulder-width ratio
Evaluation metrics (posture analysis)
Accuracy
Mean Average Precision (MAP)
Percentage of Correct key points (PCK)
Normalized Mean Error (NME)
Joint Angular Error
Focus on primarily on neck inclination angles of individuals from positive to negative degrees.
Limitation
1) Focus on the neck inclinations than other postural deviations가 문제점이였으나, full-body deviations 로 인해 보완. small sample size가 문제
2) ceratin age range of participant pool이 문제
