A Robust Lane Tracking and Curvature Estimation System in Real Time for Forward Curved Roads

• Main Authors: Yu-jung Yuan, 袁有容
• Other Authors: Chin-Shyurng Fahn
• Format: Others
• Language: en_US
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/54240367615369306957

碩士 === 國立臺灣科技大學 === 資訊工程系 === 102 === In the vigorous development of Intelligent Transportation Systems (ITS), intelligent vehicle technology is the focus of development. Not just the emphasis on safety driving assistant system, more and more academia and industry attention to the developed technology of the resolution of the discomfort by taking a car. We can through obtaining the curvature of the forward lane, while the angle of the car seat to make the appropriate adjustments to offset the centrifugal force. Intelligent vehicle technology for a wide range of development, especially for lane detection is a fundamental and important technology. In this paper, we propose a based on machine vision lane detection and tracking system in real-time. And estimate the curvature of the forward lane from our lane model. The use of Inverse Perspective Mapping (IPM) convert images to the world coordinate system, and using mean shift with the Hough transform to detect lane markings, then the lane tracking with Kalman filter. After the parabola fitting, we can get the curvature of forward lane. We carry on the experiments for the continuous curve in different environments; our proposed method can effectively detect the curved lane. The detection rate is 99.4% in multi-curves suburban environment, and the detection rate of 94.5% in the state of switch lanes in the freeway environment. The error value is less than 0.01 between our estimated curvature of the forward lane and the actual curvature. The overall performance is 23 to 30 frames per second.