GPS and Inertial Sensor Enhancements for Vision-based Highway Lane Tracking
Type of DegreeThesis
Electrical and Computer Engineering
MetadataShow full item record
For the past decade much research in the Intelligent Transportation Systems (ITS) community has been devoted to the topic of lane departure warning (LDW). A significant portion of highway fatalities each year are attributed to vehicle lane departure. Many automobile manufacturers are developing advanced driver assistance systems, many of which include subsystems that help prevent un-intended lane departure. A consistent approach among these systems is to alert the driver when an un-intended lane departure is predicted. To predict a possible lane departure, a vision system mounted on the vehi- cle detects the lane markings on the road and determines the vehicle’s orientation and position with respect to the detected lane lines. These vision-based systems suffer from performance limitations that are brought forth by environmental constraints. Therefore, it is desirable to add support from additional sensors to compensate when the vision system loses its ability to perform lane departure warning. The first goal of this research is to present current methods of vision-based LDW systems and to explore methods of sensor enhancement to assist the vision system. Sec- ond, several image processing and computer vision algorithms will be implemented as demonstration of how they could be used in a vision-based LDW system. Finally, the main goal of this research is to develop a method using additional sensors such as GPS and inertial sensors to enhance vision-based lane detection. The combination of GPS and the vision system together with a high accuracy lane- level map will allow a more robust highway lane tracking system. Kalman filtering is employed to incorporate inertial sensor inputs and measurements from the GPS receiver and vision system to estimate the vehicle states relative to highway lane tracking. Vehicle lateral offset, lateral velocity and heading angle are estimated to provide lane tracking when the vision-based lateral offset measurement fails.