基于自适应边缘配准的非结构化道路检测

摘要

为了解决非结构化道路区域分割过程中复杂背景与目标的多峰分布问题,提出了一种适用于双阈值分割的改进Otsu方法.该方法借助于灰度直方图的峰值搜索解决目标与背景方差差异较大导致的Otsu误分割问题,并利用Otsu边缘与加权Canny边缘的配准提高道路区域分割的精度.基于配准后加权Canny边缘的结果,采用自适应蒙特卡罗方法提高道路边界识别的精度,通过自适应粒子规模选取和观测模型采样机制克服传统蒙特卡罗方法的粒子退化问题.不同场景下的非结构化道路识别实验表明,该方法能够有效克服道路缺损、光影、照度变化等不利因素的影响,同时能够满足智能车辆视觉导航的实时性要求.%An improved Otsu thresholding method applicable to double-threshold segmentation for unstructured roads is proposed in this paper to resolve the multi-peak problem in both complicated background and target. This method remedies the mis-segmentation of the Otsu thresholding method caused by the diversity of between-class variances and refines the segmentation precision by matching the proposed Otsu edges with the weighted Canny edges. An adaptive Monte Carlo method based on the weighted Canny edges which has been re-evaluated by Otsu edges is also proposed to improve the precision of the road boundary and overcome the particle degradation by adaptive selection of particle size and the scheme of particle sampling in perception model. The results of the experiments on unstructured road detection in different scenes indicate that the method can not only overcome the negative influences from road flaw, changes of illumination, sunlight and shadow, but also meet the requirement of real-time visual navigation for intelligent vehicle.