Overview
- The AndroCon system interprets nine low-level GPS parameters, including Doppler shift, signal power, and multipath interference, to infer context.
- In a year-long study across roughly 40,000 square kilometres and multiple phones, it reached up to 99% accuracy for surroundings and over 87% for activity recognition.
- The framework identified states such as sitting, standing, lying down, and settings such as metro travel, air travel, parks, and crowded spaces.
- Using only GPS patterns and user trajectories, AndroCon sketched indoor floor plans, locating rooms, staircases, and elevators with errors under four metres.
- The work, led by MTech student Soham Nag with Prof. Smruti R. Sarangi and published in ACM Transactions on Sensor Networks, warns that any app granted precise location could infer sensitive information.