Tactile 센서 및 딥러닝 모델 기반 새로운 신체 압력중심점(CoP) 추정 시스템 개발 (Development of Novel Body Center of Pressure Estimation System Based on Tactile Sensor and Deep Learning Models)

서지정보

· 발행기관 : 한국로봇학회
· 수록지 정보 : 로봇학회 논문지 / 20권 / 1호 / 69 ~ 76페이지
· 저자명 : 백재현, 최윤호, 김경중, 이호수

초록

The Center of Pressure (CoP) is utilized as an essential indicator for assessing the body’s balance. CoP reflects the state of balance and is important in evaluating balance ability and predicting fall risk. Existing systems are too expensive, less accurate in dynamic conditions, or have limited measurement ability, which is for only the foot’s CoP without fully reflecting the overall body balance. Thus, this study proposes a novel system using Tactile sensors and a deep learning model for less cost and accurately estimating dynamic body CoP. The performance of the suggested CNN-Bi-LSTM model was compared with existing foot CoP estimation models, CNN-LSTM and Bi-LSTM. Model performance was validated using the Leave-One-Out Cross-Validation (LOOCV) method and evaluated with Root-Mean-Squared Error (RMSE) and R² coefficient. The experimental results showed that the CNN-Bi-LSTM model achieved the best performance, with an average RMSE of 7.09 mm in the ML direction and 4.69 mm in the AP direction, and an average R² of 0.99. In comparison, the CNN-LSTM and Bi-LSTM models recorded RMSE values of 11.59 mm and 25.52 mm in the ML direction, and 8.81 mm and 10.90 mm in the AP direction, respectively. Additionally, the RMSE difference value ​​between ML (medio-lateral) and AP (Antero-posterior) was shown to be smaller compared to previous studies on estimating the foot CoP. This result highlights the effectiveness of the CNN-Bi-LSTM model in capturing both spatial and temporal features, surpassing traditional methods and previous models in dynamic conditions. Future research will focus on expanding the system and conducting clinical trials for gait CoP analysis.

영어초록

The Center of Pressure (CoP) is utilized as an essential indicator for assessing the body’s balance. CoP reflects the state of balance and is important in evaluating balance ability and predicting fall risk. Existing systems are too expensive, less accurate in dynamic conditions, or have limited measurement ability, which is for only the foot’s CoP without fully reflecting the overall body balance. Thus, this study proposes a novel system using Tactile sensors and a deep learning model for less cost and accurately estimating dynamic body CoP. The performance of the suggested CNN-Bi-LSTM model was compared with existing foot CoP estimation models, CNN-LSTM and Bi-LSTM. Model performance was validated using the Leave-One-Out Cross-Validation (LOOCV) method and evaluated with Root-Mean-Squared Error (RMSE) and R² coefficient. The experimental results showed that the CNN-Bi-LSTM model achieved the best performance, with an average RMSE of 7.09 mm in the ML direction and 4.69 mm in the AP direction, and an average R² of 0.99. In comparison, the CNN-LSTM and Bi-LSTM models recorded RMSE values of 11.59 mm and 25.52 mm in the ML direction, and 8.81 mm and 10.90 mm in the AP direction, respectively. Additionally, the RMSE difference value ​​between ML (medio-lateral) and AP (Antero-posterior) was shown to be smaller compared to previous studies on estimating the foot CoP. This result highlights the effectiveness of the CNN-Bi-LSTM model in capturing both spatial and temporal features, surpassing traditional methods and previous models in dynamic conditions. Future research will focus on expanding the system and conducting clinical trials for gait CoP analysis.

참고자료

· 없음