A Novel FSVM with PSO for gait phase detection based on elastic pressure sensing insole system

• Published in: International journal of intelligent robotics and applications Online Vol. 8; no. 3; pp. 596 - 608
• Main Authors: Lv, Pingping, Zhang, Chi, Yi, Feng, Yuan, Ting, Li, Shupei, Zhang, Meitong
• Format: Journal Article
• Language: English
• Published: Singapore Springer Nature Singapore 01-09-2024; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Computer Science; Control; Discriminant analysis; Electronics and Microelectronics; Exoskeletons; Gait; Human body; Human mechanics; Insoles; Instrumentation; Kinematics; Lightweight; Machine learning; Machines; Manufacturing; Mechatronics; Neural networks; Particle swarm optimization; Physiology; Plantar pressure; Processes; Regular Paper; Robotics; Robots; Sensors; Support vector machines; User Interfaces and Human Computer Interaction; Walking; Weight reduction; Gait phase; Plantar pressure measurement system; Lower limb exoskeleton; Support vector machine; Radius-margin error; Particle swarm optimization
• ISSN: 2366-5971; 2366-598X
• DOI: 10.1007/s41315-024-00334-1

The precise gait phase detection with lightweight equipment under variable conditions is crucial for low limb exoskeleton robots. Therefore, the kinematics and dynamics information are investigated. In this paper, a novel radius-margin-based support vector machine (SVM) model with particle swarm optimization (PSO) in feature space called PSO-FSVM is proposed for gait phase detection. The proposed method addresses the dual objectives of maximizing margin while minimizing radius, employing PSO to fine-tune the parameters of the FSVM. This enhancement significantly bolsters the classification accuracy of the SVM. For the measurement of gait features with a lightweight sensor system, the plantar pressure insoles equipped with flexible and elastic sensors are designed. To evaluate the effectiveness of our method, we conducted comparative experiments, pitting the proposed PSO-FSVM against other support vector machine variants, across four treadmill speeds. The experimental results indicate that the proposed method achieves an accuracy of over 98% at four different speeds indoors. Furthermore, the proposed method is compared with other algorithms (SVM, k-nearest neighbor (KNN), adaptive boosting (AdaBoost), and quadratic discriminant analysis (QDA)) under outdoor experiments. The experimental results demonstrate that the average recognition accuracy of this method reaches 96.13% under variable speed conditions, with an average accuracy of 98.06% under slow walking conditions, surpassing the performance of the above four algorithms.