Effect of dynamic feature for human activity recognition using smartphone sensors

Human activity recognition (HAR) from time series sensor data collected by low cost inertial sensors attached to small portable devices like smartphones are increasingly gaining attention in various fields especially for health care, medical, millitary and security applications. The need for efficient time series data analysis for recognition of human activities has enhanced research efforts in this area. For correct recognition of human activities, efficient feature selection from the time series data is important. In this work an approach for dynamic feature extraction from time series human activity data is proposed and classification results with dynamic features and static features are compared. The efficiency of dynamic features over static features are noted by simulation experiments with benchmark data set with different classifiers available in machine learning domain. Experiments are also done with convolutional neural networks(CNN) for activity recognition using extracted dynamic features. It is found that CNN provides better recognition accuracy for dynamic activity recognition with dynamic features compared to conventional classifiers such as multilayer perceptron (MLP), support vector machine(SVM) or k-nearest neighbour(KNN) though it takes higher computational time and memory resources.