Collecting Vehicle Trajectory Information by Smartphones when GPS Signal is Lost Mecit Cetin, PhD Department of Civil and Environmental Engineering Tamer Nadeem, PhD Computer Science Ilyas Ustun, Abdulla Alasaadi, Olcay Sahin, Matthew Orensky
Introduction Smartphones market share As of Dec 2013, 65% have it Smartphones equipped with various sensors GPS, accelerometer, gyroscope, compass, proximity sensor, and ambient light sensor, etc. Sensor data used in: Activity classification (sitting, walking, running, biking) Travel mode identification (car, bus, light rail, riding a bike, walking) 2
Objective Without using GPS detect whether the vehicle is in motion or stopped estimate speed Why avoid GPS? Low accuracy of GPS in urban areas with tall buildings, tunnels, covered areas, Low precision of GPS localization, High-power consumption when the GPS is in use 3
Project Description Funded by TranLIVE UTC (University Transportation Center) Develop an app to estimate fuel consumption and CO 2 emissions for a multi-modal trip Provide feedback to the driver 4
Potential Applications Traffic monitoring Signal timing optimization Fuel consumption and emission predictions Estimating vehicle trajectories while in a tunnel 5
Data Collection within Tunnels 6
Data Collection System 7
Data Collection Android App Acceleration, Gyroscope, Magnetometer sensor data GPS latitude, longitude, and speed OBD speed (if available) GPS and OBD speeds are used for model development and testing 8
GoGreen App (in Development) 9
Database View 10
Collected Data 11
Three-axis Accelerometer Data 12
Raw Acceleration Data in Y and Z 13
Std Deviation of Y and Z 14
Range Range Calculated over 1-sec Data Orientation invariant data 2.5 Driver 16_2014-06-12_Trip 1 dodge caliber, LGE Nexus 5 Motion percentage, Overlap: 0.5, Window: 1 secs, 0.59 2 1.5 1 0.5 Motion Standstill 8.6 8.8 9 9.2 9.4 9.6 9.8 10 10.2 10.4 Total Acceleration 15
Hidden Markov Model The X s are hidden states (vehicle being stationary or in motion). The Y s are visible observations (acceleration data). 16
Speed (mph) Sample Data Toyota Prius 70 60 50 40 30 Driver 2_2014-06-16_Trip 1, GPS speed used, Overlap: 0.5, Window: 1 secs, toyota prius, samsung GT-N7000 Motion percentage 0.74, HMM trained at and the emission vector created at: Range: 0.5 stop point move point Threshold X squared std dev mean range max gps speed threshold pred HMM pred 20 10 0 00:00 02:04 04:08 06:12 08:16 10:20 12:24 14:28 16:32 18:36 20:41 22:45 index 17
Speed (mph) Sample Data - Dodge Caliber 70 60 50 40 30 Driver 16_2014-06-12_Trip 1, GPS speed used, Overlap: 0.5, Window: 1 secs, dodge caliber, LGE Nexus 5 Motion percentage 0.59, HMM trained at and the emission vector created at: Range: 0.6 stop point move point Threshold X squared std dev mean range max gps speed threshold pred HMM pred 20 10 0 00:00 01:05 02:11 03:17 04:23 05:29 06:34 07:40 08:46 09:52 10:58 12:03 index 18
Likely states Speed (mph) 60 Detecting Motion and Stops Acura RL Speed 40 20 0 Likely states Observation Actual stop point Actual move point -0.1-2.1-0.3-1.0-1.8 0.1-0.8-0.3 0.9-0.8 0.5 0.7 0.2-0.4 00:00 02:15 04:31 06:47 09:02 11:18 13:34 15:49 18:05 20:21 22:37 24:52 Time (mm:ss) Device: Samsung Note I Vehicle: Acura RL 2000 19
Likely states Speed (mph) Detecting Motion and Stops Toyota Prius 50 40 Speed 30 20 10 0 Likely states Observation Actual stop point Actual move point 0.1-1.1 0.1-0.8 0.1-0.1-0.3-0.5-0.6-0.4-4.2-0.3-0.4-0.6-0.7 00:00 02:04 04:08 06:12 08:17 10:21 12:25 14:30 16:34 18:38 20:43 22:47 Time (mm:ss) Device: Samsung Note I Vehicle: Toyota Prius 2007 20
Likely states Speed (mph) Detecting Motion and Stops Dodge Caliber 80 Speed 60 40 20 0 Likely states Observation Actual stop point Actual move point -0.9-0.5-1.2-0.1-0.5-0.8 00:00 01:06 02:12 03:18 04:24 05:30 06:36 07:42 08:48 09:54 11:00 12:06 Time (mm:ss) Device: LGE Nexus 5 Vehicle: Dodge Caliber 2011 21
Mean Acceleration (m/s 2 ) Speed Estimation Accelerometer data Noisy and biased Bias varies from phone to phone Mean Acceleration Readings ( over 11s) 10.6 10.512 10.4 10.2 10.233 10.219 10.0 9.8 9.796 9.717 9.6 9.499 9.4 9.2 9.0 8.8 +z -z +y -y +x -x Orientation (axis facing up / opposing gravity) 22
Speed (mph) Speed Estimation Before Calibration 500 400 File: 1379013576840ALLdata Resample freq: 10 Hz Est speed OBD speed GPS speed 300 200 100 0-100 00:00 02:26 04:53 07:19 09:46 12:13 14:39 17:06 19:32 21:59 24:26 26:52 Time (Min) 23
Speed (mph) Speed Estimation After Calibration 80 70 60 50 File: 1379013576840ALLdata Resample freq: 10 Hz Est speed OBD speed GPS speed 40 30 20 10 0-10 00:00 02:26 04:53 07:19 09:46 12:13 14:39 17:06 19:32 21:59 24:26 26:52 Time (Min) 24
00:00.0 00:14.1 00:28.1 00:41.9 00:56.3 01:10.8 01:25.1 01:39.1 01:53.2 02:07.4 02:22.0 02:36.1 02:50.1 03:04.5 03:18.8 03:32.7 03:46.8 04:00.9 04:14.9 04:29.2 Speed (mph) Speed Estimation Using PCA 40 35 30 25 GPS Speed PCA Speed 20 15 10 5 0 Time 25
Conclusions Developed an App and DB to collect high resolution data from smartphones The number of stops and their times can be estimated with relatively good accuracy Speed estimation with calibration is feasible for obtaining acceptable results Continue the research to develop more robust and accurate models 26