Activity recognition plays an essential role in bridging the gap between the low-level sensor data and the high-level applications in ambient-assisted living systems. With the aim to obtain satisfactory recognition rate and adapt to various application scenarios, a variety of sensors have been exploited, among which, smartphone-embedded inertial sensors are widely applied due to its convenience, low cost, and intrusiveness. In this paper, we explore the power of triaxial accelerometer and gyroscope built-in a smartphone in recognizing human physical activities in situations, where they are used simultaneously or separately. A novel feature selection approach is then proposed in order to select a subset of discriminant features, construct an online activity recognizer with better generalization ability, and reduce the smartphone power consumption. Experimental results on a publicly available data set show that the fusion of both accelerometer and gyroscope data contributes to obtain better recognition performance than that of using single source data, and that the proposed feature selector outperforms three other comparative approaches in terms of four performance measures. In addition, great improvement in time performance can be achieved with an effective feature selector, indicating the way of power saving and its applicability to real-world activity recognition.

The inherent complexity of human physical activities makes it difficult to accurately recognize activities with wearable sensors. To this end, this paper proposes a hierarchical activity recognition framework and two different feature selection methods to improve the recognition performance. Specifically, according to the characteristics of human activities, predefined activities of interest are organized into a hierarchical tree structure, where each internal node represents different groups of activities and each leaf node represents a specific activity label. Then, the proposed feature selection methods are appropriately integrated to optimize the feature space of each node. Finally, we train corresponding classifiers to distinguish different activity groups and to classify a new unseen sample into one of the leaf-nodes in a top-down fashion to predict its activity label. To evaluate the performance of the proposed framework and feature selection methods, we conduct extensive comparative experiments on publicly available datasets and analyze the model complexity. Experimental results show that the proposed method reduces the dimensionality of original feature space and contributes to enhancement of the overall recognition accuracy. In addition, for feature selection, returning multiple activity-specific feature subsets generally outperforms the case of returning a common subset of features for all activities.