适用于右转辅助的非机动车道检测
Bicycle lane detection suited for right turn assistance
刘翔鹏 1彭雨琳 1童德中 2陆玮 1王丹宁1
作者信息
- 1. 上海师范大学 信息与机电工程学院,上海 201418
- 2. 加利福尼亚大学洛杉矶分校 机械与航天工程学院,洛杉矶 90095
- 折叠
摘要
为避免传统车辆或无人驾驶车辆在右转时与快速行驶的自行车发生碰撞的风险,需要在转弯前稳定地检测出非机动车道,并实时定位骑车者.为实现该目标,通过调整初始代价及代价聚合,完成更为精确的非机动车道图像立体匹配;基于改进的Hough变换算法,对路面上的非机动车道线进行识别和提取;之后结合立体视觉来获取非机动车道的精确三维坐标.实验结果表明:所提方法可以在大部分场景下检测出不同类型的非机动车道,且具有较高的精确度,适用于广义非机动车道检测,对于先进驾驶辅助系统或无人驾驶车辆实现全道路场景识别起到了积极的辅助作用.
Abstract
In order to avoid the risk of collisions between traditional vehicles or unmanned vehicles and fast-moving bicycles when turning right,it was necessary to detect the bicycle lane stably and locate the cyclist in real time.By adjusting the initial cost and cost aggregation,more accurate three-dimensional matching of bicycle lane images was completed.Based on the improved Hough transform algorithm,the bicycle lane lines on the road were identified and extracted;then 3D vision technique was combined to obtain the accurate 3D coordinates of the bicycle lane.Experimental results showed that the proposed method was able to detect different types of bicycle lanes in most scenarios with high accuracy,accordingly it was suitable for generalized bicycle lane detection.The proposed algorithm would play a positive auxiliary role in realizing scene understanding of all-road scenarios for advanced driving assistance systems or unmanned vehicles.
关键词
非机动车道检测/驾驶辅助/立体匹配/Hough变换/车道线拟合Key words
bicycle lane detection/driving assistance/stereo matching/Hough transform/lane line fitting引用本文复制引用
基金项目
上海师范大学一般科研项目(SK202123)
出版年
2024
NETL
NSTL
万方数据