A low power and high accuracy MEMS sensor based activity recognition algorithm

Wearable sensors and smart phones have been used in human activity recognitions and can achieve relative high accuracy however the power consumption is also high. In this paper, we propose an activity recognition approach that can achieve high accuracy with low power consumption. Two strategies have been applied to reduce the power consumption. The first strategy is using the hierarchical support vector machine classification algorithm to reduce the computational complexity. The second strategy is to reduce the sensor data sampling rates. Data collected from sensors in low sampling rate were processed using a wider time window for the feature extraction. The experiment results show that the average recognition accuracy of human activities (sitting, standing, walking, and running) in 1 Hz sampling rate can reach 98.50%. It indicates that the proposed approach can effectively extend the battery lifetime while maintaining high prediction accuracy in activity recognition.