Mobile Sensing for Efficient and Accurate Office Analytics

I delivered this presentation at the Computer Laboratory (University of Cambridge) during the mini conference for research students.

Transcript

1. Mobile Sensing for Efficient and Accurate Office Analytics Alessandro Montanari alessandro.montanari@cl.cam.ac.uk Supervisor: Prof. Cecilia Mascolo cecilia.mascolo@cl.cam.ac.uk
2. What is office analytics? Sensing Analysis Action
3. Informal team communication is synonymous of performance and efficiency
4. Observations are not cost effective and do not scale People forget and give socially desirable answers
5. My Vision Research new techniques to collect data with existing mobile devices Benefits • Always with the user • Familiarity with devices • More privacy control
6. Experimental Wearable Platform Device • Bluetooth Low Energy • 3-Axis accelerometer • Vibrator motor • SD card for data recording Data Collection • Fine grained proximity between people • Desk level localisation • Motion status (walking or no-movement) Electronics in 3D printed box Principle of operation • Simultaneous transmission and scanning of beacons
7. Commercial Wearable Devices Samsung Gear S2 (Tizen Wearable) • Relatively low receive rates • Energy efficient Samsung Gear Live (Android Wear) • High receive rates • Power hungry Simultaneous Transmission and Scanning • Available since June/September 2015
8. Deployment Ground Floor Basement Location • Spacelab Ltd. (London) • 17 static beacons for localisation • 2 smartphones as base stations • Charging station for devices Participants • 25 people recruited for 4 weeks • Mainly architects and designers • Dynamic working style (hot desks) • Ground truth: 19 hours of observations
9. 1 5 10 Window Size (s) True Positive Rate 0.0 0.2 0.4 0.6 0.8 1.0 Custom Device Tizen Balanced Tizen Low Power Android Wear Low Power Proximity Detection Results Custom Device • Benefits from complete freedom in parameters settings • ~20 hours of battery life Gear S2 (Tizen) • Good trade off between accuracy and power consumption • ~17 hours of battery life Gear Live (Android) • Lower accuracy due to limited control over the parameters • ~13 hours of battery life
10. Inter-team and Inter-level Mixing Teams are well mixed and interact frequently More contacts among people in the lower levels of the hierarchy rather than in the higher levels
11. Locations of Contacts • Most of the contacts in open space workstations and corridors • Contacts in meeting rooms mainly among high level roles (directors) • Similar amount of contacts in the kitchen due to lunch together
12. What I’ll do next • Investigation of sensor fusion techniques across mobile devices • Inertial sensors and WiFi for orientation detection • Microphone for conversation monitoring • Spatio-temporal modelling of human behaviour • Where, how and under which conditions different interactions flourish • Improvement of existing frameworks to support architectural research • Include dynamic aspects of a building (e.g. where people sit, popular areas)
13. Thank you! Alessandro Montanari alessandro.montanari@cl.cam.ac.uk http://www.cl.cam.ac.uk/~am2266/