Department of Civil, Environmental, and Geo-Engineering (CEGE) Battery-Efficient Location Change Detection for Smartphone-Based Travel Data Collection: A Wi-Fi Fingerprint Approach (#16-2090) Chen-Fu Liao, Yingling Fan, Gediminas Adomavicius, and Julian Wolfson 2 3 4 1 Dept. of Civil, Environmental, and Geo-Engineering Hubert H. Humphrey School of Public Affairs Carlson School of Management School of Public Health 1 2 3 4 Abstract Methodology Summary and Conclusion With the advance of smartphone technology and prevalence of smartphone ownership, numerous apps are being made available in the market to enhance the quality of our life. In transportation, a plethora of apps have been developed by researchers to collect trip and activity data for travel mode detection, trip experience, and traveler behavior analysis. However, one fundamental concern about the smartphone-based data collection apps is the battery power consumption. When the battery power depletes rapidly from continuous operation of multiple sensors (especially the GPS receiver) on the smartphone, the requirement (and inconvenience) of frequent battery recharge often outweighs the potential benefit of using those apps. In this work, we present an approach to conserving smartphone’s battery power using the information from the surrounding Wi-Fi networks, i.e., using the Wi-Fi signal fingerprint technique. We use two indices, Jaccard and normalized weighted signal level change (NWSLC), using the Wi-Fi signal fingerprint technique to detect a location change. Both measures are computed based on the information about the received Wi-Fi signal strength, obtaining which is substantially less demanding on the smartphone battery than obtaining continuous GPS information. The proposed measures were implemented on a test app for verification and validation. Experiments were conducted at 10 different indoor locations (representing potential trip origins) to determine the appropriate numeric thresholds for location change detection. The results suggest that proposed Wi-Fi fingerprint methodology can reliably detect the change of a person’s location prior to resuming the GPS service for trip data collection. Objective The objective of this study is to develop a Wi-Fi signal fingerprint methodology for detect location change when a person initiates a trip. The GPS service will be turned off to conserve battery power when no motion is detected by the accelerometers for a period of time. In areas when Wi-Fi signals are available, the GPS service will not be turned back on until a location change or beginning of a trip is detected by the Wi-Fi fingerprint technique. Smartphone Sensors Data Analysis and Results Introduction We build upon the SmarTrAC app (recently developed by our research group). This smartphone app is designed to incorporate smartphone sensing with advanced statistical and machine learning techniques to automatically detect, identify, and summarize attributes of a traveler’s daily trips and activities. We learned that the battery power consumption rate for the GPS sensor is significantly larger than that of the accelerometer (i.e., motion sensor). We found that phones consumed minimal power (1.0% - 1.1% per hour) when GPS sampling was turned off, and that phones consumed 4.9% - 7.0% power per hour when GPS sampling was turned on. Table 1 Comparison of Accelerometers on Different Smartphones Smartphone Brand Samsung Galaxy S5 Samsung Galaxy S4 HTC One Sony Xperia Z MEMS Manufacturer Invensense ST Microelectronics Bosch Bosch Model MPU-6500 K330 (LSM330DLC) BMA-250 BMA-250 Type 6-Axis 6-Axis 3-Axis 3-Axis Acceleration Tri-Axis accelerometer programmable full scale range of ±2g, ±4g, ±8g and ±16g Tri-Axis accelerometer programmable full scale range of ±2g, ±4g, ±8g and ±16g Tri-Axis accelerometer, programmable full scale range of ±2g, ±4g, ±8g and ±16g Tri-Axis accelerometer, programmable full scale range of ±2g, ±4g, ±8g and ±16g Angular Rate Tri-Axis angular rate sensor (gyro) with a sensitivity up to 131 LSBs/dps and a full- scale range of ±250, ±500, ±1000, and ±2000dps Tri-Axis angular rate sensor (gyro) with a full-scale range of ±250, ±500, and ±2000dps NA NA Linear Acceleration Sensor Range 0...19.613 (±2g) 0...19.613 (±2g) 0...39.227 (±4g) 0...156.96 (±16g) Resolution 0.001 (0.003%) 0.001 (0.003%) 0.038 (0.098%) 20 (12.742%) Power 0.25mA 0.25 mA 0.1 mA 0.003 mA Web Site invensense.com www.st.com ae-bst.resource.bosch.com We used a Jaccard index and a normalized weighted signal level change (NWSLC) index for detecting location changes. We also develop a general-purpose battery power conservation methodology for travel data collection apps to determine when to suspend/resume the GPS sensor on the smartphone. Wi-Fi Fingerprinting Technique Jaccard Index (or Similarity Coefficient) The Jaccard index, or similarity coefficient, of two sets S 1 and S 2 is defined as: 1 , 2 = 1 ∩ 2 1 ∪ 2 ∈ 0,1 (1) S i = { BSSID | (BSSID, RSSI) F i , SignalLevel (RSSI, 5) ≥ 2 }, for i = 1, 2. Normalized Weighted Signal Level Change (NWSLC) Index = 1 =1 × − (2) Where, is the NWSLC index, n is the number of intersection samples (i.e., n = | F 1 F 2 |), N is the total number of signal levels (i.e., N = 5 in our case), is the signal level of reference AP i (from scan F 1 ), is the signal level of current AP i (from scan F 2 ). Site ID Test Site Description* Initial # of APs AVG Jaccard Index AVG NWSLC Index From To From To 1 From a work office to an elevator in the building 7 1.00 0.09 0.00 1.34 2 From an elevator to the terrace outside a building 75 0.72 0.01 0.22 0.91 3 From an indoor atrium to the building’s east exit 152 0.56 0.01 0.40 1.29 4 From a student study area to a building’s east exit 73 0.71 0.01 0.38 2.58 5 From a lobby area to the building’s east exit 181 0.70 0.08 0.32 1.32 6 From a café area to the building’s north exit 112 0.92 0.00 0.15 1.65 7 From inside a Starbucks store to its north exit 37 0.69 0.07 0.47 1.24 8 From a hotel lobby to its west exit 31 0.76 0.03 0.32 1.08 9 From a home office to the neighbor’s driveway 14 1.00 0.09 0.00 0.91 10 From the lobby of an alumni center to its N. exit 158 0.91 0.02 0.19 1.59 *Notes Stay at each indoor location for about 1 min. AVG 0.80 0.04 0.25 1.39 Travel to a location outside the building or store about 30-50 meters away SD 0.15 0.04 0.16 0.49 Flowchart We introduced two indices based on monitoring the Wi-Fi signal fingerprint for detecting location changes, the Jaccard index and the normalized weighted signal level change (NWSLC) index. The Jaccard index measures the similarity of two sample sets. The normalized weighted signal level change (NWSLC) is another index that describes the changes of a network signature by weighting the signal level change with respect to its previous signal strength and then taking the normalized average. Both measures based on the received Wi-Fi signal strength were implemented on a smartphone app for validating the proposed location change detection logics. The results from experiments at 10 indoor locations indicate that proposed Wi-Fi fingerprinting methodology can reliably detect location change before turning on GPS sensor for trip data collection. Battery Power Consumption Transportation Mode Identification