Automated recognition of human movement states using body acceleration signals

Automated recognition of human activity states has many advantages, e.g., applications in the smart home environment for the monitoring of physical activity levels, detection of accidental falls in the older adults in the home environment or assessment of the recovery phase of patients living independently at home. In this paper, we describe an accelerometer-based system to recognize three activity states, e.g., steady state gait or walking, sitting and simulated sudden accidental falls. The recorded 3D movement accelerations of the trunk were processed using wavelets, and the features were extracted for recognition of movement states through the use of a fuzzy inference system. The system was trained and tested using 58 different data segments representing the three states. Cross-validation test results indicated an overall recognition accuracy by the machine classifier to be 89.7% with an ROC area of 0.83. The results suggest good potential for the system to be applied for various situations involving activity monitoring as well as gait and posture recognition. Further tests are required using various population groups.