权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

The Ego-Lane Detection under rainy condition

雨天条件下的自我车道检测

基本信息

批准号：
09650485
负责人：
OZAWA Shinji
金额：
$ 1.6万
依托单位：
Keio University
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (C)
财政年份：
1997
资助国家：
日本
起止时间：
1997 至 1998
项目状态：
已结题

来源：
https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-09650485/
关键词：
Rainy Weather Lane-Marker Template Matching Moving Image Processing モデル

项目摘要

Generation of Lane-marker candidate-imageThe Lane-Marker of the road in both fine and rainy weather proved to have the feature that the intensity of the image became peak around the Lane-Marker area by examining the profile of road surface and the area. By using the feature, we tried to extract the Lane-Marker's candidate-regions which have high probability to be white lines, where input sequential images taken from the camera installed on the ruuning car.Then, by speeding up our system with the image processor, we could also adopt the information of previous frames. As a result, our system could generate more hopefull Lane-Marker candidate-image. Robust Recognition of Ego-LaneWe considered a road boundary as an arc model, and we extracted the road boundary by our template matching process between the Lane-Marker candidate-image. The measurement of road's radius of curvature and the position and moving direction of the car could be obtained, that is, the Ego-Lane could be recognized, where the direction of the car was parallel with the road's line.Then, we improved the above method by adding a line model to road model, etc. At first, the straight line model were used as a template since we assumed the Lane-Maker was approaching to a straight line near the car, and the position and moving direction of the car could be obtained. Next, by using these information and applying the arc model, the road's radius of curvature can be obtained.As the necessary model parameters were increased, the reseaching space of the template matching process became large. But we could reduce the space by using our method in order which was expalined in the previous statements, and ccould get a better capability to recognize the Ego-Lane.Meanwhile, the motion of wiper in rainy weather is also considered. Since the motion disturbs a good scene for detection of Lane-Marker, we proposed a method to eliminate the scene with wiper so as not to reduce the detection capability.

车道标线候选图像的生成通过检测路面和区域的轮廓，证明了晴天和雨天道路的车道标线具有在车道标线区域周围图像强度达到峰值的特征。利用该特征提取车道标记的候选区域，这些区域很可能是白线，输入的序列图像是安装在车牌上的摄像头拍摄的，然后通过使用图像处理器来加速系统，也可以采用以前帧的信息。因此，我们的系统可以生成更多的希望车道标记候选图像。EGo-Lanewe的稳健识别将道路边界看作一个弧形模型，通过对Lane-Marker候选图像之间的模板匹配过程提取道路边界。在此基础上，通过在道路模型中加入直线模型等方法对上述方法进行了改进，首先以直线模型为模板，假设造车道人正在接近汽车附近的一条直线，从而得到汽车的位置和移动方向。然后，利用这些信息，应用圆弧模型得到道路的曲率半径，随着所需模型参数的增加，模板匹配过程的搜索空间变大。但是，我们可以通过使用我们的方法来按顺序缩小空间，从而获得更好的识别自我车道的能力，同时还考虑了雨天雨刷的运动。由于运动干扰了车道标记物检测的良好场景，我们提出了一种在不降低检测能力的前提下去除带有雨刷的场景的方法。