高效公路车道检测：直线模型与K-means聚类法

66 浏览量更新于2024-08-27 1 收藏 1.52MB PDF 举报

本文主要探讨了一种基于直线模型和K-means聚类的高效且鲁棒的高速公路车道检测算法。研究团队由Jinyu Liu、Lu Lou、Darong Huang、Yu Zheng和Wang Xia组成，分别来自中国重庆交通大学信息科学与工程学院和重庆交通职业技术学院轨道交通工程系。该方法首先利用霍夫变换(Hough Transform)来识别公路鸟瞰视图中的车道线。霍夫变换是一种用于检测图像中特定形状（如直线）的方法，它通过累积投票的方式找到所有可能的直线候选，提高了对车道线的精确识别能力。接着，通过K-means聚类算法将所有提取出的车道线进行分类，这是一种无监督学习技术，通过迭代优化将数据集分为多个类别，使得内部相似度高而不同类别之间的差异大。在聚类后，算法会选择每个类别中最具有代表性的车道线作为最终结果。这种策略有助于减少车辆和护栏等道路特征对检测结果的干扰，提高算法的抗噪性能。实验证明，这种方法在处理车辆和障碍物的影响时表现出色，能够达到90%的正确率，对于高级驾驶辅助系统(ADAS)的车道保持功能尤其有益。研究指出，随着中国高速公路事故率逐年上升，车道检测技术在交通事故预防方面具有重要意义。通过结合直线模型和K-means聚类，研究人员开发出一种能够有效提升驾驶安全性和行车便利性的智能解决方案。本文的关键字包括K-means聚类、反向透视映射（Inverse Perspective Mapping，IPM）以及车道检测，这些都是实现自动驾驶和智能交通系统的核心要素。这项研究不仅提供了一个实用的车道检测方法，还展示了如何在复杂交通环境中利用先进的计算机视觉技术来增强驾驶员辅助系统的能力，对于推动智能交通领域的研究和发展具有积极的推动作用。

Lane Detection Based on Straight Line Model and K-Means Clustering

Jinyu Liu

, Lu Lou

, Darong Huang

, Yu Zheng

, Wang Xia

1. College of Information Science and Engineering, Chongqing Jiaotong University, Chongqing 400074, China

E-mail: hcx1978@163.com

2. Department of Rail Transit Engineering, Chongqing Vocational College of Transportation, Chongqing 402247, China

E-mail: zhengyu523@163.com

Abstract: This paper presents an effective and robust algorithm to detect the lanes in highway. It uses Hough Transform to fit the

lane line of top view of the road and extracts the most representative lane line in each category after clustering all the lines, which

is then followed by a post-processing step. The results show that this algorithm can effectively reduce the disturbance of vehicles

and guardrails to achieve 90% correct rate.

Key Words: K-Means Clustering, Inverse Perspective Mapping, Lane Detection

1 Introduction

The incidence of highway traffic accidents has increased

year by year, which is a threat to public health in China. Lane

detection is a useful technology of Advanced Driver

Assistance System (ADAS) which may help reducing the

driving fatalities and has already attracted a lot of attention.

The lanes of highway are usually white or yellow lines,

and it can be divided into many types including single solid

line, double solid line, dotted line. Between the lanes of the

ramp, there are also labeled a few traffic signs, such as speed

limit, steering, road signs. The complex road environments

bring great challenges for the lane detection in ADAS. In

recent years, many breakthroughs have been made in lane

detection for highway, urban road and suburban roads. Lane

detection algorithms are mainly divided into three categories:

the first type is based on the model-based lane detection.

Paper [3][4][5][6] detected the lanes based on line model

with better real-time performance, but it did not overcome

the disturbance of road signs; in [9], a camera imaging model

was built and the lanes were scanned twice in conjunction

with the edge contribution function to overcome the

disturbance of non-lane markings; in [7][8], a Random

Sample Consensus (RANSAC) curve model was set up to fit

the lane line; in order to improve the detection rate, paper [10]

used the direction-variable Haar feature to extract the edge

feature points and fitted them by hyperbola model; in [11], a

generalized curvilinear lane model was established firstly,

and then the parameters of the model were solved by

adaptive selection of random Hough Transform and taboo

search according to the parameters of the model, which

improved the real-time performance of the algorithm. The

second type is feature-based lane detection. After Inverse

Perspective Mapping (IPM), paper [12] used parallel

features of lane preforming the enhancement of lane and then

extracted; in [13], the candidate lane markings are marked by

the color characteristics of lanes and the main lane markings

are found by clustering; paper [14] used template matching

to extract the feature points of the lane edge; in [15], the

lane-line characteristic area is found through the

This work is supported by National Nature Science Foundation under

Grant 61573076, Scientific and Technological Research Program of

Chongqing Municipal Education(KJ1755492)

characteristics of the boundary line between the road and the

non-road, and the lane line is fitted by straight line or curve

model; in [16], the invariant moment features in the image

are extracted, and the Support Vector Machine (SVM) is

used to model the invariant moments of the lane line to detect

the lane line; in [17], the color feature was used to extract the

lane area, and the edge features were used to test the lane line,

finally, RANSAC was used to lane-lane curve fitting. In

method of feature-based, some noise edge points can have a

significant impact on the test. The third type is the lane

detection based on deep learning in recent years; for example,

in [18], an improved Boosting algorithm was used to train

the classification function on the edge and the gray level of

the image to judge whether each point is the edge point of the

lane line; paper [19] used Extreme Learning Machine (ELM)

for image enhancement and lane detection; in order to solve

the interference of fences, trees and crossroads, paper [20]

used Convolutional Neural Network (CNN) to suppress the

noise in images and then used RANSAC algorithm to fit lanes;

paper [21] used two-time transfer learning in SegNet

network for end-to-end lane segmentation and achieved

good results; in [22], a multitask network structure based on

Recurrent Neural Networks (RNN) and CNN was designed

to detect the lane line, and the effect of vehicle blockage and

lane wear on lane line detection was well solved. Lane

detections based on Deep Learning increase the recognition

rate, but they require the higher hardware support.

End-to-end detection of Deep Learning is not conducive to

the decision-making of automatic driving intermediate

process.

This paper focuses on the highway lane detection. In

structured road environments in highway, the lane markers

are clear and the scene interference is relatively weak. The

lane detection algorithm based on features or models can

achieve the better detection results; paper [12] pointed out

that different pixels in the image are used to represent the

same position in the field of view due to the influence of

perspective and it first used IPM to eliminate the influence of

perspective to detect the lane line; paper [7] combined with

RANSAC and IPM algorithm to detect urban lanes to solve

the impact of sun glare, other road signs, different pavement

materials on lane detection, and has better real-time and

robustness; in [5], a combination of structure and gradient

method was proposed to extract candidate lane lines, and

527

2018 IEEE 7th Data Driven Control and Learning Systems Conference

May 25-27, 2018, Enshi, Hubei Province, China

DDCLS'18

下载后可阅读完整内容，剩余5页未读，立即下载

weixin_38722184

粉丝: 5
资源: 899

高效公路车道检测：直线模型与K-means聚类法

K均值聚类用于点集聚类或基于点集聚类的分割等当面

直接聚类法进行点的聚类

基于聚类的建筑物直线特征提取算法 (2012年)

对于文本数据，情感分析预测模型和K-means聚类如何层级使用

k-means聚类分析事故发生率 k-means聚类分析事故发生率 k-means聚类分析事故发生率代码

基于流形距离的k-means聚类算法

k-means聚类算法python

k-means聚类优点

K-means聚类优点

k-means聚类算法python包

最新资源