IODetector: A Generic Service for Indoor Outdoor Detection Pengfei Zhou†, Yuanqing Zheng†, Zhenjiang Li†, Mo Li†, and Guobin Shen‡ †Nanyang Technological University, Singapore ‡Microsoft Research Asia, Beijing, China Sensys 2012 Presenter: SY
Goal Define indoor/outdoor – High accuracy – Prompt response – Energy efficiency
How Mobile phone – Light sensor – Cellular RSSI – Magnetic field signal Detection Aggregation
Applications GPS management Wifi scanning Context aware computing Activity recognition
Outline System design – Light detector – Cellular detector – Magnetism detector Aggregation Evaluation Case Study Conclusion
System Overview
Light Sensor – Key Observation Reading from mobile phones (discrete)
Light Sensor – Key Observation Reading from TelosB Rotation in outdoor
Light Sensor – Detection Process Query proximity sensor for readings If > threshold s1, it is outdoor/semi- outdoor with high confidence If it is daytime, it is indoor with high confidence Else, not sure 1.Check another threshold s2 1.If s2 < L < s1 indoor, C L = (s1-L)/s1 2.if L < s2 outdoor, C L = (s2-L)/s2 Else, not sure 1.Check another threshold s2 1.If s2 < L < s1 indoor, C L = (s1-L)/s1 2.if L < s2 outdoor, C L = (s2-L)/s2
Cellular Signal – Key Observation Signal from current active cell tower – Handover problem – Corner effect
Cellular Signal – All Towers
Cellular Detector Use all visible cell towers n number of visible cell towers N+(t) -> number of towers whose RSS increases more than v N-(t) -> number of towers whose RSS decreases more than v N0(t) -> number of towers whose RSS change between +/-v n number of visible cell towers N+(t) -> number of towers whose RSS increases more than v N-(t) -> number of towers whose RSS decreases more than v N0(t) -> number of towers whose RSS change between +/-v
Magnetic Detector Variance Empirical threshold a = 18 Compute variance over t = 10s Confidence level Cm = t/10 Variance Empirical threshold a = 18 Compute variance over t = 10s Confidence level Cm = t/10
Pros And Cons Fast and accurate Indoor vs outdoor/semi-outdoor Not always available Widely available Indoor vs outdoor/semi-outdoor Require sufficient # of towers Indoor/semi-outdoor vs outdoor Available only when moving Light Detector Cellular Detector Magnetism Detector
Aggregated IODetector Stateless IODetector Find the highest confidence level
State Changes Current state is usually related to previous states
Stateful IODetector First order HMM Transition and emission probabilities are determined by training experiments Transition probabilities
Aggregated IODetector Stateless – Estimate based on instant detection results – Not that stable Stateful – Infers current environment considering previous state – Robust to noises – Needs continuous detection Use accelerometer to trigger detection
Experiment Setup Mobile phones – Samsung Galaxy S2 i9100, HTC Desire S, and HTC Sensation G14 Sensor nodes – TelosB – Connects to mobile phone (for light sensor) Environments
Sub-detector Performance
Aggregated IODetector
Energy Consumption Negligible
Case Study – Adaptive GPS
GPS Performance
IODetector-Augmented GPS
Energy Consumption
Conclusion Use available sensors on mobile phone Lightweight – Low energy consumption Pretty good accuracy Arguments in case study is probably weak