Human daily activity recognition by fusing accelerometer and multi-lead ECG data

Human daily activity recognition has gained much attention since it has a wide range of applications. In this paper, we propose a novel scheme for recognizing human daily activity by fusing multiple wearable sensors, i.e., accelerometer and multi-lead ECG. Firstly, both time and frequency domain features are extracted from raw sensor data. In order to alleviate the computation complexity of subsequent process, the dimensions of feature vectors would be sharply reduced by performing linear discriminant analysis (LDA). Then, the reduced feature vectors are classified by relevance vector machines (RVM). Finally, considering different sensors and leads would provide complementary information about the human activity, the individual classification results are fused at the decision level to improve the overall recognition performance. Experimental results show that if seven leads of ECG and accelerometer are fused, we can even achieve recognition accuracy as high as 99.57%. Furthermore, the proposed scheme has great potential in real-time applications due to its strong ability in feature dimensionality reduction, simple classifier structure, and perfect recognition performance.