Ultra Low Complexity Algorithm for Lane Departure Warning System

• Main Authors: Kuan-Chieh Wang, 王冠傑
• Other Authors: 吳崇賓
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/57070007140078944283

碩士 === 國立中興大學 === 電機工程學系所 === 103 === The thesis proposed a fast linear block-based lane detection and departure warning system. Based on the analysis of the distribution of lane markings, the lane markings in near distance are extracted that they are straight lines. A region of interest (ROI) is determined, and the lane markings in the region are strengthened to be suitable in different weather conditions. For SLDA, we extract the colors of lanes in a determined region of interest (ROI), and the result is partitioned into non-overlapped blocks. The Sobel edge filter is applied to compute the angles of blocks. According to the angles, the blocks are classified into 6 groups that a fixed gradient is defined to each group. An adaptive standard deviation of intercept is proposed to filter out redundant blocks. After filtering, the block number of each group is used to determine left and right candidate lanes. Two mean points of the two chosen groups are calculated and the vanishing points of two lanes are also determined. With the mean points and the vanishing points, the descriptive functions of left and right lanes are determined. The gradient of left and right lanes’ descriptive linear functions are used for the decision of lane departure warning. If the gradient is larger than or equal to the pre-set threshold, the lane departure warning signal is sent. As for FLDA, to reduce the computation, we proposed two arrays to compute the horizontal and horizontal gradient of a block that the block angle and block gradient are also obtained. Based on the driving conditions, the blocks are classified into 6 groups by their angles. The two groups with maximum numbers are chosen as the left and right lane markings. To describe the lane markings with linear function formula, the average gradient and gradient of each group are computed. According to the gradients of the lane markings and the pre-set thresholds, the decision of warning signal is determined. The experimental results show that the average detection rate in SLDA achieves 91.6%, the average processing speed in 40.19ms. And the average detection rate in FLDA raised to 96.41%, the average processing speed can reach 3.71ms. Furthermore, FLDA is more adaptive than SLDA in various environments.