Date of Award
Doctor of Philosophy (PhD)
Modeling Simul & Visual Engineering
Vehicle trajectories are an important source of information for estimating traffic flow characteristics. Lately, several studies have focused on identifying a vehicle’s trajectory in traffic network using data from mobile devices. However, these studies predominantly employed GPS coordinate information for tracking a vehicle’s speed and position in the transportation network. Considering the known limitations of GPS, such as, connectivity issues at urban canyons and underpasses, low precision of localization, high power consumption of device while GPS is in use, this research focuses on developing alternate methods for identifying a vehicle’s trajectory at an intersection and at a urban grid network using sensor data other than GPS in order to minimize GPS dependency. In particular, accelerometer and gyroscope data collected using smartphone’s inertial sensors, and speed data collected using an on-board diagnostics (OBD) device, are utilized to develop algorithms for maneuver (i.e., left/right turn and through), trip direction, and trajectory identification.
Different algorithms using threshold of gyroscope and magnetometer readings, and machine learning techniques such as k-medoids clustering and dynamic time warping are developed for maneuver identification and their accuracy is tested on collected field data. It is found that, clustering based on maximum and minimum value of gyroscope readings is effective for maneuver identification. For trip direction identification at an intersection, two different methods are developed and tested. The first method utilizes accelerometer, gyroscope and OBD speed data, and the 2nd method employs magnetometer and acceleration data. The results demonstrate that the developed method using accelerometer, gyroscope and OBD speed data are effective in identifying a vehicle’s direction. An effective algorithm is developed using OBD speed information, maneuver and trip direction identification algorithms to identify vehicle’s trajectory at a grid network. Techniques for noise removal and orientation correction to transfer the raw data from phone’s local coordinate to global coordinate system are also demonstrated.
Overall, this research eliminates the need for continuous GPS connectivity for trajectory identification. This research can be incorporated in methods developed by researchers to estimate traffic flow, delays, and queue lengths at intersections. This information can lead to better signal timings, travel recommendations, and traffic updates.
Ahmed, Umana. "Algorithms for Constructing Vehicle Trajectories in Urban Networks Using Inertial Sensors Data from Mobile Devices" (2017). Doctor of Philosophy (PhD), dissertation, Modeling Simul & Visual Engineering, Old Dominion University, https://digitalcommons.odu.edu/msve_etds/8