基于多层级联视觉显著性模型的肇事车辆锁定方法  

Localization of Causing-traffic-trouble Vehicle with Multi-level Cascaded Visual Attention Model

作　　者：柴桢亮 臧笛[2] 

机构地区：[1]同济大学计算机科学与技术系,上海201804 [2]同济大学嵌入式系统与服务计算教育部重点实验室,上海200092

出　　处：《计算机科学》2015年第4期285-291,共7页Computer Science

基　　金：国家自然科学基金(61103071);教育部博士学科点新教师基金(20110072120065);留学回国人员科研启动基金;2012年科技部国际合作专项(2012DFG11580)资助

摘　　要：肇事车辆的锁定是智能交通系统中一个十分重要的问题,因此针对肇事车辆的锁定,提出了一种基于多层级联视觉注意模型的肇事车辆匹配方法。在模型的每一层中,基于传统视觉注意模型的思想,通过生成显著图的方式提取车辆的一个显著性特征,如颜色、车标,并将其与肇事车辆进行匹配,过滤掉特征不相似的车辆,经过多次显著性特征提取和匹配,最终获得唯一的肇事车辆。实验结果表明,该模型可以准确地从车辆数据库中锁定肇事车辆,且对光照变化和噪声有较强的鲁棒性。The localization of causing-traffic-trouble vehicle is one of the most key problems for intelligent transportation system（ITS）.This paper proposed a multi-level cascaded visual attention model to localize the causing-traffic-trouble vehicle.In each level of the proposed model,one significant feature of the vehicle such as color or vehicle logo is extracted and compared to the vehicle which has caused an accident.Then vehicles that have no similar features can be filtered.By performing feature extraction and feature comparison for several times,only the causing-traffic-trouble vehicle will be left behind.The experimental results demonstrate that the proposed approach is able to locate the causing-traffictrouble vehicle accurately and is robust to luminance and noise.

关 键 词：肇事车辆 车辆匹配 计算机视觉 视觉注意模型 车牌识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]