基于视觉的高速公路匝道场景理解

项目名称： 基于视觉的高速公路匝道场景理解

项目编号： No.61472053

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 其他

项目作者： 房斌

作者单位： 重庆大学

项目金额： 82万元

中文摘要： 高速公路匝道口的正确识别是无人车是否能够有效完成指定任务的关键之一。本项目围绕基于视觉的高速公路匝道场景理解这一课题展开，研究仅依靠视觉的自然环境下的高速公路匝道场景分析和理解:(1)场景图像预处理：建立基于曲面曲率的无参考图像质量评价模型，降低或消除低质量图像对场景理解能力的影响;(2)感兴趣区域提取：融合自顶向下和自底向上的视觉注意机制，构建任务相关的感兴趣区域检测模型;(3)目标对象分割：综合图形边缘和区域信息，构造新的B样条水平集分割模型，确定待处理对象区域;(4)导向箭头识别：构造对象的特征统一描述并形成统计模板，利用拓扑结构信息优化特征点集匹配，并识别判断;(5)匝道场景理解：利用多帧导向箭头的联合信息准确判断车辆前方高速公路是否有匝道。通过本项目的实施，为车辆在高速公路匝道场景自主行驶的实时路况分析提供技术支撑。

中文关键词： 智能车；图像质量判断；视觉注意融合机制；B样条水平集；特征点集拓扑编码

英文摘要： To correctly identify freeway ramp is one of the key tasks that a unmanned-vehicle is able to effectively conduct its assigned task. This project focuses on Vision-based scene understanding of freeway ramp, devoting to analyze and understand the visual scenes of freeway ramp: ( 1 )Pre-process scene image : establish a no-reference image quality index based on surface curvature to reduce and eliminate low-quality scene image's adverse impact on scene understanding; ( 2 ) Extract the region of interest : integrate the top-down and bottom-up visual attention mechanism to construct task-related detection model of interest region ; ( 3 ) Segment the potential target: combine edge and regional information, construct a new B-spline level set segmentation model to determine the potential target area; ( 4 ) Recognize the directional arrows: format the unified description of objects' characters, construct their statistical templates, optimize the feature point set matching by topology information and recognize the directional arrows ; ( 5 )Identify the ramp : employ the multi-frame information of directional arrows to identify the freeway ramp. The implementation of this project will provide technical support of real-time freeway scene analysis for the unmanned-vehicle drive.

英文关键词： Unmanned-vehicle;Image quality assessment;Integral visual attention mechanism;B-spline level set;Feature-set's topological code