Revised paper – Dec 1, 2010. MobiCon: Mobile Context Monitoring Platform for Sensor-Rich Dynamic Environments 1Seungwoo Kang, 2S.S. Iyengar, 1Youngki Lee, 1Chulhong Min, 1Younghyun Ju, 1Taiwoo Park, 1Jinwon Lee, 3Yunseok Rhee, 1Junehwa Song. 1Department of Computer Science, Korean Advanced Institute of Science and Technology (KAIST), South Korea. 2Department of Computer Science, Louisiana State University, Baton Rouge, USA. 3 Electronics & Information Eng., Hankuk University of Foreign Studies, South Korea. ABSTRACT We present a new mobile context monitoring platform to support emerging pervasive applications in a Personal Area Network (PAN)-scale dynamic mobile computing environment. Composed of a mobile device and many wearable and space-embedded sensors, the PAN-scale computing environment will constitute an important part of future pervasive spaces technology. The monitoring platform presented in this paper supports a number of context-aware applications to simultaneously run and share highly scarce and dynamic resources. Our new system develops a novel translation-based approach for efficient and effective monitoring architecture for the environment. We implement and test a prototype system on multiple mobile devices with a diverse set of sensors. Example applications are also developed based on the implemented system design. Experimental results show that the system achieves a superior level of scalability and energy efficiency. The processing time can be reduced by orders of magnitude compared to what the present technology offers, and data transmission can be reduced by the order of 50 percent. 1. INTRODUCTION Smart mobile devices will be the central gateway for personal services in the emerging pervasive environment (Figure 1). They will enable a lot of personal context-aware applications, forming a personal sensor network with a number of diverse sensor devices, placed over human body or in surrounding spaces. Diverse sensors act as the useful tool for the applications to acquire users’ contexts1, i.e., current status of an individual or surrounding situation that she/he faces into, without their intervention [42]. Table 1 summarizes example contexts, sensors, and related applications. For example, diverse physical contexts such as heart rate can be recognized from biomedical sensors such as ECG, GSR, and BVP sensors, and gait from accelerometers and gyroscopes. Also, environmental status can be obtained from light/ temperature/dust sensors, GPS, RFIDs in daily 1 While one of important contextual information is profile-based context such as age, gender, profession, contacts, and SNS buddy lists which is rather static, this paper focuses on sensor-enabled context which is more dynamic status information derived from sensing data. surrounding spaces [5][18][19][23][24] and the network of such sensors [26][27]. Diverse contexts enable mobile applications to proactively provide users with personal services customized to their situation. Such context-aware applications increasingly emerge in practice and in diverse research domains, e.g., healthcare [3][29], elderly support [25], dietary monitoring [2], daily life assistant [4], and sports training. Proactively providing context-aware services often needs continuous monitoring of users’ context. The context monitoring is a process of continuous detection of context changes. It requires continuously collecting data from diverse sensors, processing the data to derive context, and keeping track of changes of users’ context. This semantics is different from conventional context recognition, which identifies the current context. Once a change is identified, it is not necessary to recognize the context redundantly as long as it remains unchanged. The context monitoring often involves multi-step complex operations such as feature extraction and context recognition distributed across the mobile and sensor devices at the same time. However, it is quite difficult for individual applications to perform complicated context monitoring process on their own, especially over the shared devices with highly limited resources. To effectively support context monitoring, a common monitoring platform is essential. Figure 1. PAN-scale Sensor-rich Mobile EnvironmentRevised paper – Dec 1, 2010. Table 1. Example contexts and applications Challenges and Existing Approaches Enabling mobile context monitoring involves complex, multilateral challenges spanning over different research and technical issues. Applications require different types of contexts in different degrees of awareness and accuracy. Users have different requirements and preferences for the services as well as privacy concern. Efficient resource utilization and management also becomes more important issue for continuous context monitoring. To enable diverse context-aware application with ease and efficiency, in particular, infrastructural supports including a platform are required. Such challenges have been addressed in diverse research domains evolving independently with different aspects and emphasis. We introduce such research efforts in four major areas. Artificial intelligence and machine learning. Many research efforts have been done to recognize user contexts automatically, especially focusing on recognition accuracy and complex context extraction from sophisticated sensors such as camera, e.g., arm tracking and face gesture detection. They have addressed a variety of challenges across multi-stages in context recognition, e.g., pre-processing, feature extraction, segmentation, and recognition algorithms. They try to extract the most effective features for accurate recognition (e.g., MFCC for audio [49], statistical or FFT features for activity [5]) from raw sensing data (e.g., camera, microphone, biomedical and motion sensors). From continuous feature data, data sections corresponding to valid context values are automatically segmented without user intervention (e.g., probability-based segmentation [50]). Most importantly,