单目视觉车道偏离预警系统的开发  

Design of lane depature warning system with monocular vision

作　　者：吕柯岩[1,2,3] 朱明[1,3] 杨迪[1,2,3] 

机构地区：[1]中国科学院航空光学成像与测量重点实验室(中国科学院长春光学精密机械与物理研究所),长春130033 [2]中国科学院研究生院,北京100039 [3]中国科学院长春光学精密机械与物理研究所,长春130033

出　　处：《计算机应用》2012年第A02期196-198,241,共4页journal of Computer Applications

基　　金：中国科学院二期创新基本资助项目(C05T022)

摘　　要：车道偏离预警系统的主要任务是当驾驶员在驾驶过程中非意识发生车道偏离时,采用预警的方式来保障完成汽车的安全。采用基于机器视觉的方法,设计了一款用单个摄像机采集道路信息,用TMS320DM642作为主处理器对采集到图像进行预处理并提取车道线,然后结合汽车CAN总线传输的汽车当前的状态信息进行预警决策。硬件方面,双片SDRAM对图像进行高速缓存以解决视频数据量大的问题,TVP5150对采集到的图像进行解码为后续图像处理提供数据,SAA7121对图像数据进行编码以实现实时视频显示。算法方面,首先对图预处理后,依据道路特征设定感兴趣区域,使用改进Hough变换提取车道线的线形特征。结果证明,本系统采用单目可以降低成本,算法可以很好地移植,能够提前2.5 s进行偏离预警,达到实时性要求。Lane departure warning system＇ s main task is that the system makes warning for the driver when the car occurs departure unconsciously to protect the car. Based on machine vision, this paper designed a new system, which uses one camera to gather the image information of the road. TMS320DM642 was used as the main processor which deals with the information including prctreating the image and finding the lanes, and then it was judged whether warning should be made or not associating with the car＇ s state through the car CAN Bus. In hardware, dual-SDRAM was used as memory to store lots of video information, and TVP5150 was used for decoding the image, meanwhile, SAA7121 was used for coding the image. The Region Of Interest （ROI） in the image was set after pretreating the image, and the lanes were extracted using the improved Hough transform. The results turn out that the system cost less with monocular, and the algorithm can be transplanted easily, at the same time, the system can pre-warn in 2.5 seconds.

关 键 词：车道偏离预警系统 机器视觉 汽车CAN总线 感兴趣区域 改进HOUGH变换 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]