计算机科学 ›› 2010, Vol. 37 ›› Issue (1): 290-293.

• 图形图像及体系结构 • 上一篇    下一篇

基于视觉动态模型的道路检测算法研究

石磊,金忠,杨静宇,王煜   

  1. (南京理工大学计算机科学与技术学院 南京210094);(泰达有线电视网络有限公司 天津300457)
  • 出版日期:2018-12-01 发布日期:2018-12-01
  • 基金资助:
    本文受国家自然科学基金(No.60705020, No. 60873151)资助。

Robust Road Detection Based on Vision Dynamic Modeling

SHI Lei,JIN Zhong,YANG Jing-yu,WANG Yu   

  • Online:2018-12-01 Published:2018-12-01

摘要: 道路的检测与识别是自主导航车辆视觉感知研究的重要问题之一。在阴影、光照不均匀、边缘信息模糊等情况下,道路的识别与环境的理解变得尤为困难。为了提高道路识别中杭干扰和鲁棒性能力,提出了一种基于视觉动态模型的道路检测算法。该算法利用计算机视觉方法对摄像机观测系统进行建模,引入了直线变形模型描述道路环境的几何结构。基于道路图像的连续性建立对车辆运动状态的动态预测模型,并利用粒子滤波算法对状态模型参数进行预测,以达到对道路边缘的跟踪。同时,引入了似然概率作为评价道路图像与结构模型的匹配程度。通过对大量实际路面的检测,证明了该方法的有效性和实用性。

关键词: 主动导航车辆,直线变形模型,动态模型,粒子滤波,似然概率

Abstract: Road detection and identification are one of the essential problems for vision understanding in autonomous vehide systems. However, it becomes relatively harder with the surrounding conditions such as shadows, asymmetric illumination, and blurry edge information. To enhance the robustness and interference immunity, a Vision Dynamic Mode ling based road detection algorithm was proposed. This method structures camera observation systems on computer vision, and introduces deformablc line model for road geometry description. Based on continuity a dynamic model for the vehicle motion was constructed, which incorporates Particle Filtering into road tracking by parameters estimation and prediction. Besides,likelihood probability combined with road model was utilized to evaluate the fitness with sampled images. Experiments with road images from CMU lab and our system proved the effectiveness and applicability of this algorithm.

Key words: Autonomous vehicle systems,Deformable line model,Dynamic modeling,Particle filtering,Likelihood probability

No related articles found!
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!