Human Activity Recognition using Smartphone Sensors with Context Filtering

In recent times, application of Ambient Intelligent services, e.g. smart home, remote monitoring and assisted healthcare, the use of smart phones for the recognition of human activities has become a topic of high interest. Simple activities like sitting, running, walking can be recognized easily but semi-complex activities like ascending and descending stairs, slow running, jogging, fast running etc. are often difficult to recognize accurately. We aim to reduce the error rate of recognizing these kinds of activities by applying Dynamic Time Warping (DTW) algorithm and introducing context filtering. We used heart rate data and barometric pressure sensor data as elements of context filtering. We used a steady state as template and matched every activity with this steady state. To get optimum threshold values, we applied K Nearest Neighbor (KNN) algorithm on the score of DTW. After primarily classifying activities, we used the context filtering approach to further recognize activities by removing confusions. After completion of our study, we have seen that accuracy level has increased significantly for differentiating similar kinds of activities. Overall, our novel approach of applying DTW algorithm and applying context filtering shows considerable performance improvements at a low cost. Keywords-Human Activity Recognition (HAR), context filtering,