By following this approach, the system is able to classify both ambulatory as well as transportation contexts, while still achieving low power consumption. Ganetespib FDA The overall architecture of the proposed solution is presented in Figure 1.Figure 1.Overall architecture of the proposed system��Context Recognizer.As described in Figure 1, for the overall architecture, we used Gaussian Mixture Model (GMM) for the acceleration data classification and Hidden Markov Model (HMM) for the audio classification. Before modeling and classifying acceleration data, a prior process including feature extraction and selection generates bunch of features to be used for a classification. In order to use multiple dimensions of features, mixture model which is suitable for representing multiple distributions of collected data is chosen.
Other classification techniques such as Gaussian Process are more appropriate for considering small number of variables or features. For the audio classification, we used HMM algorithm for training and testing audio data because the module needs to be classify only two activities��bus and subway��and requires running on a smartphone in real-time. There are other audio classification algorithms such as Conditional Random Field and Support Vector Machine, but our approach using HMM is lighter than other algorithms and it fits in classifying similar audio data both collected from bus and subway.2.?Related WorksThe high availability of smartphones with built-in sensors (accelerometer, gyroscope, GPS, Wi-Fi, etc.) is highly advantageous to the research area of context recognition.
In [3,6,7], a smartphone accelerometer was used to recognize user movement contexts such as walking and running; in [5,8], the author utilized audio data to classify acoustic environments. The authors of [4,9,11] showed that GPS can be used to recognize Batimastat transportation routines. However, we must note that those works merely exploited a particular sensor instead of combining the strength of multiple sensors. To the best of our knowledge, [2] is one of the first works to combine accelerometer and audio classification; the author demonstrated that the combination of audio helps improve the accuracy of recognizing user activities.In [13], the authors designed and implemented both an audio classifier and accelerometer classifier using audio and accelerometer sensors.
The modules are similar to our work but the approaches to recognize contexts are different. In their system, each classifier can recognize only one specific context��the accelerometer classifier recognizes human behaviors such as sitting, standing, walking and running, on the other hand, the audio http://www.selleckchem.com/products/Cisplatin.html classifier’s purpose is to determine whether a person is in a conversation or not��but our proposed system utilizes both classifiers and other sensors together for classifying contexts as described in Figure 1.