GPS-Aided Lane Marking Detection and Vehicle Positioning

E. Broshears

Abstract:	Autonomous control of mobile robots and vehicles is an area currently under highly-active research, especially in the field of vision navigation. Image processing and image navigation are actively being researched because there are many complications, which means there are many different ways to go about solving these. This research uses only a single monocular camera and a low-cost GPS receiver to give a more accurate and frequently updated vehicle position solution. GPS receivers usually have an update rate of about 1 Hz, while some can go as high as 10 Hz or even higher depending on the quality of the receiver. Video cameras naturally have a higher capture frequency, since the human eye can perceive motion smoothly at about 30 frames per second. Some low-cost cameras can even have frame rates of over 100 frames per second. This would allow updating the vehicle's global position solution much more frequency than just using the GPS receiver alone. Doing this could have implications in controlling multiple vehicles in a close vicinity on the highway in an electronic stability control or collision avoidance scenario. In the 1970s, the Air Force planned to implement a global positioning system that used satellites strategically placed GPS positioning is used widely in navigation because of it's many advantages. The errors that affect the GPS position solution are zero mean, and the derived velocity is very accurate. However, the applications the Air Force intended to use the GPS for were scenarios with extremely high visibility of the sky, such as on the ocean or up in the air. Because of this, GPS outages occur frequently in urban environments due to the trees and buildings that block the signals to the satellites to the receiver. Sensor fusion is one alternative to alleviating this problem. A variety of sensors are viable alternatives, such as radars, lidars, or inertial measurement units. This research integrates a monocular camera into the GPS positioning to improve the accuracy by detecting lane markings with well-surveyed GPS positions known a priori. The position accuracy of a low-cost, single-frequency GPS receiver is typically about 10 meters. This is fine for navigation if you just need to know on which road the vehicle is traveling, but certain applications require more accuracy, such as collision avoidance or object detection. This research uses a combined GPS receiver and single monocular camera vision navigation system. Vision navigation is especially difficult because of all the variables. Lighting and weather can be an extreme detriment to any object detection algorithm, and also like objects can come if different shapes, colors, and sizes. However, the lane markings in the middle of the road have a greater consistency in size and shape as well as disparity from the road on which it is painted. A lane marking detection algorithm was created to find lane markings on the road and calculate the distance and angle to that lane marking. With this knowledge, depending on the resolution and quality of the camera, the position solution of the car can be calculated to less than 1 meter accuracy. To detect the lane markings, first a canny edge detection is ran on the images. This finds the gradients in the image, and isolates blocks that are surrounded by this gradient. By changing the tolerance of the gradients, more or less objects can be detected. Because lane markings generally have straight sides and consistent widths, an algorithm is created to check the width and variance of the width of each object. If the object passes the test, it is deemed a lane marking. At a test track, the front of each lane marking was surveyed using GPS. Since each lane marking is about 15 meters apart, as long as the GPS position of the vehicle is within these bounds, it can be assumed which lane marking is located in front of it in view of the camera. Once the lane marking has been detected and analyzed, the distance and location to the front of the lane marking can be calculated, bringing the vehicle's position accuracy on the road down to less than 1 meter. Using cameras to aid GPS positions has a big advantage in update rate. If the camera is recording at 30 frames per second, the vehicle's position can be updated 30 times between each GPS measurement. This has drastic implications for highway driving or collision avoidance that GPS wouldn't be able to detect in time. In conclusion, a low-cost, single-frequency GPS receiver can be coupled with a single monocular camera to improve the update rate and position accuracy of a vehicle on a road with known lane marking global positions. This has applications in either a single vehicle scenario, or in a relative positioning scenario involving a cluster of vehicles. As long as the camera is properly calibrated and the lane markings properly surveyed, the quality of the camera does not to be high end, and is an easy addition to a GPS positioning solution.
Published in:	Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013) September 16 - 20, 2013 Nashville Convention Center, Nashville, Tennessee Nashville, TN
Pages:	1325 - 1328
Cite this article:	Broshears, E., "GPS-Aided Lane Marking Detection and Vehicle Positioning," Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013), Nashville, TN, September 2013, pp. 1325-1328.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In