Activity recognition combines hard sensing and soft sensing inputs to a computer to determine what an individual is doing. Hard sensing inputs include:
- Accelerometers (measuring relative motion)
- Location sensing
- Ambient light
- Ambient audio
Soft sensing inputs include:
- Device activity
- Social networking actions
- Calendar data
The goal of this research is to use these elements to intelligently guide, instruct, encourage, inform, and otherwise support each user’s activities in a personalized way. The computer becomes more of an intelligent assistant to your everyday activities.
Practical applications of this form of context-aware computing include determining a person’s level of physical activity to help physicians treat lifestyle-dependent diseases, such as diabetes and heart disease. Individuals usually don’t accurately report, for example, the amount of exercise they get each week. With tracking and guidance provided by a mobile computer, both the physician and the patient have real data and analytical tools to more realistically assess goals and monitor health improvements.
Going beyond physical activity to higher-level semantics of activities can help improve everyday communication or enable introspection. Imagine grabbing your phone and, before dialing your wife, you see that she is in a meeting. How would that change your behavior? Or, imagine being able to look at a summary of how you spent your time over the last year—in the same way you do today with your financial information. What would this type of analysis reveal? These examples suggest the ways these kinds of capabilities can be used to bring new experiences to users.
Activity recognition can be combined with other elements, such as gesture recognition, social proximity, and emotional classifiers to develop a wide range of highly personalized applications. Intel Labs is actively working in all these areas to advance context-aware computing and provide a platform for innovation.
Identify areas where activity recognition data can be fused with other forms of context-aware data to create innovative mobile computing applications.
- Improve the fidelity of activity recognition algorithms and create personalized models.
- Develop context algorithms to be used as representative workloads, guiding the design of future platforms.
- Discover techniques to improve the responsiveness and usefulness of computers in supporting the lifestyles and activities of computer users.
Healey, Jennifer, Lama Nachman, Sushmita Subramanian, Junaith Shahabdeen, and Margaret Morris. Out of the Lab and into the Fray: Towards Modeling Emotion in Everyday Life. In Lecture Notes in Computer Science, Volume 6030/2010, 156–173, DOI: 10.1007/978-3-642-12654-3_10, 2010.
Maurer, Uwe, Asim Smailagic, Daniel P. Siewiorek, and Michael Disher. Activity Recognition and Monitoring Using Multiple Sensors on Different Body Positions. From Proceedings of the International Workshop on Wearable and Implantable Body Sensor Networks, IEEE Computer Society, 2006.
Nachman, Lama, Amit Baxi, Sangeeta Bhattacharya, Vivek Darera, Piyush Deshpande, Nagaraju Kodalapura, Vincent Mageshkumar, Satish Rath, Junaith Shahabdeen, and Raviraja Acharya. Jog Falls: A Pervasive Healthcare Platform for Diabetes Management. Pervasive Computing, 8th International Conference, Pervasive 2010, Helsinki, Finland.
Patterson, Donald J., Dieter Fox, Henry Kautz, and Matthai Philipose. Fine-Grained Activity Recognition by Aggregating Abstract Object Usage. International Symposium on Wearable Computers, 2005.