CCTV 촬영각이 영상 기반 자동 OWAS 평가 정확도에 미치는 영향

초록

Objective: This study examined the effects of CCTV horizontal and vertical viewing angles and identified camera configurations that provide relatively stable performance in video-based automated OWAS assessment by comparison with motion capture data. Background: Video-based human pose estimation has emerged as a promising alternative to conventional observational methods for musculoskeletal risk assessment. However, the influence of CCTV viewing angles on posture recognition accuracy has not been sufficiently validated. Method: A laboratory experiment was conducted simulating a PE pipe manufacturing task. Video data were collected under five horizontal viewing angles (0~180°) and two vertical angles (30°, 60°). Video-based postures were estimated using MediaPipe, and motion capture data were used as ground truth. Performance was evaluated using joint angle errors and OWAS classification accuracy. A two-way repeatedmeasures ANOVA was applied to examine the effects of viewing angles. Results: Viewing angle had significant effects on OWAS accuracy and joint-level estimation errors. Overall, the lower vertical angle (30°) yielded more stable performance than the higher vertical angle (60°). The highest OWAS code accuracy and the lowest OWAS action category RMSE were observed at the rear low-angle condition (180~30°), whereas side-oriented and high-angle conditions generally increased errors. Conclusion: Among the tested conditions, low-angle front and rear views provided relatively stable performance for video-based automated OWAS assessment. The maximum exact OWAS code accuracy was 59.2%, supporting the use of single-view CCTV outputs for screening and comparative monitoring in conjunction with expert or sensor-based evaluation. Application: The findings provide camera-placement guidance for automated posture monitoring; low-angle front or rear views are preferred whereas elevated or sideoriented installations can be supplemented with additional cameras or calibration procedures.

키워드