Human Motion Understanding Based On Sensor Fusion

Wenxing Li,Fei Wang,Heping Chen,Jianhui Wang,Zhen Zeng,Rongjie Li

DOI: https://doi.org/10.1109/cyber.2018.8688161

2018-01-01

Abstract:At present, the identification of arm movements is mainly focused on the discrete motion mode classification. The individual arm movement signals have differences and instability. To improve the recognition rate of the surface myoelectric signals when the arm continuously moves under different assembly modes, the traditional methods are mainly based on classifier parameter adjustment or feature extraction techniques. This paper focuses on the recognition of arm motion patterns through sEMG signals and inertial sensor information in different assembly modes, and identifies and classifies different assembly patterns, then maps the motion trajectories of different assembly patterns to robots and uses kinematics to the demonstration program of the robot is realized. Based on this research, an arm motion pattern recognition based on time series analysis is proposed. The arm motion pattern is classified and identified using HMM and an improved SVM method. Two different classifiers perform complementary combination. It integrates HMM's ability to model sequential data and support vector machine's classification and recognition capabilities. It uses SVM's strong classification ability to reflect the differences among heterogeneous samples, handles the classification problem among multiple samples, and then uses the HMM process of hand movement. The continuous signal in the process is processed, modeled, and the model establishment and recognition process is completed. The comparison experiments show that the hybrid method has significantly improved the accuracy of arm motion pattern recognition compared with a single support vector machine (SVM) or a single Hidden Markov (HMM) arm motion pattern recognition: the average recognition rate is 93%. The rate is 10%-15% higher than simple SVM classification recognition. Finally, it is used for ABB YUMI robots to realize human-computer interaction by teaching the robot different movement patterns.