Gait detection of lower limb exoskeleton robot integrating visual perception and geometric features

• Published in: Intelligent service robotics Vol. 18; no. 3; pp. 529 - 551
• Main Authors: Huang, BinHao, Lv, Jian, Qiang, Ligang
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.05.2025; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Cameras; Cerebral palsy; Classification; Control; Dynamical Systems; Engineering; Exoskeletons; Gait; Gait recognition; Machine learning; Mechatronics; Medical research; Methods; Motion capture; Older people; Optimization; Original Research Paper; Parameter estimation; Patients; Plantar pressure; Pressure distribution; Rehabilitation; Robotics; Robotics and Automation; Sensors; Support vector machines; Target detection; User Interfaces and Human Computer Interaction; Vibration; Visual perception; Wearable technology; Exoskeleton geometric feature; Gait detection; Swarm optimization algorithm; Machine learning; Visual perception
• ISSN: 1861-2776; 1861-2784
• DOI: 10.1007/s11370-025-00598-x

Efficient extraction of gait features is essential for enhancing human–machine collaboration in wearable devices such as lower limb exoskeletons. This study combines the Yolov7-tiny target detection model with the MediaPipe framework to propose a method for extracting gait information from visual data, eliminating the dependence on sensors typical of traditional gait data extraction methods. It derives three gait feature parameter calculation methods suited to lower limb geometric characteristics and classifies gait phases based on plantar pressure distribution. A support vector machine optimized by the improved giant armadillo optimization algorithm was introduced as a machine learning classifier for performance validation. Comparisons with three other optimization algorithms validated the accuracy and superiority of the gait detection method. The impact of camera spatial positioning on detection results and the variation in detection accuracy across different subjects were also analyzed, along with anti-interference testing. Experimental results show that the proposed lower limb gait detection method achieves an accuracy of up to 99.2%, meeting the precision requirements of wearable devices. Camera distance significantly influences detection results, although angle differences showed no significant effect. Detection accuracy remained consistent across subjects, and the proposed method demonstrated some resistance to spatial interference, indicating its effectiveness in performing gait detection for exoskeletons.