Junaith Ahemed Shahabdeen

IMOTE2: Serious Computation at the Edge

While the majority of sensor network application research to date has focused on applications where the controlling processor need only be capable of handling simple communications and data primitives required for low rate sampling, comparatively few researchers have targeted the other end of the spectrum; applications where there is a significant advantage to performing data processing at the edge of the network. This paper describes Imote2, an advanced sensor network node that enables such usage models, and introduces several applications and scenarios where we have found such edge processing to be a significant advantage.

Jog falls: a pervasive healthcare platform for diabetes management

This paper presents Jog Falls, an end to end system to manage diabetes that blends activity and energy expenditure monitoring, diet-logging, and analysis of health data for patients and physicians. It describes the architectural details, sensing modalities, user interface and the physicians backend portal. We show that the body wearable sensors accurately estimate the energy expenditure across a varied set of active and sedentary states through the fusion of heart rate and accelerometer data. The GUI ensures continuous engagement with the patient by showing the activity goals, current and past activity states and dietary records along with its nutritional values. The system also provides a comprehensive and unbiased view of the patients activity and food intake trends to the physician, hence increasing his/her effectiveness in coaching the patient. We conducted a user study using Jog Falls at Manipal University, a leading medical school in India. The study involved 15 participants, who used the system for 63 days. The results indicate a strong positive correlation between weight reduction and hours of use of the system.

Out of the lab and into the fray: towards modeling emotion in everyday life

We conducted a 19 participant study using a system comprised of wireless galvanic skin response (GSR), heart rate (HR), activity sensors and a mobile phone for aggregating sensor data and enabling affect logging by the user. Each participant wore the sensors daily for five days, generating approximately 900 hours of continuous data. We found that analysis of emotional events was highly dependent on correct windowing and report results on synthesized windows around annotated events. Where raters agreed on the timing and quality of the emotion we were able to recognize 85% of the high and low energy emotions and 70% of the positive and negative emotions. We also gained many insights regarding participants perception of their emotional state and the complexity of emotion in real life.