| Whole Machine and System Design |
|
|
|
|
| Design of online lane line recognition system based on machine vision |
| LI Meng |
| Department of Mechatronics Engineering, Anhui Institute of Information Engineering, Wuhu 241000, China |
|
|
|
Abstract Aiming at the problem that the current lane line recognition system is difficult to realize the online recognition of lane lines because the processing object is offline image or video files, most lane line recognition algorithms usually select a fixed region of interest (ROI) region to reduce the amount of image processing, which makes them difficult to adapt to the dynamic changes of the environment and there are different degrees of recognition errors. For this reason, an online lane line recognition system without setting region of interest was proposed based on the machine vision. Firstly, the real-time collected color road images were preprocessed through the VBAI (Vision Builder for Automation Inspection) platform to complete the grayscale, filtering and binary processing. Secondly, the image coordinate system was established, and the gray value collection lines diverging outward were constructed. The gray value at the the intersection of the gray value collection line and the lane line changed greatly. When the gray value of a point was higher than the setting gray threshold, the point was recorded as the edge sudden change point. Thirdly, all lane edge sudden change points were fitted by fit line algorithm to complete the lane line recognition. At the same time, the remote vanished point coordinate of the left and right inner lane lines and the relative deviation angle of the vehicle deviating from the lane center line were solved. When the relative deviation angle exceeded the safety thresholds of different levels, the prompt box of the system's human-computer interaction interface presented different colors for reminding or warning. Lastly, the LabVIEW was used to call the API (application programming interface) script to realize the continuous operation of image processing programs and online lane line recognition. Experimental results showed that the accuracy of the proposed online lane line recognition system was over 98.41%, the measurement error of relative deviation angle was less than 0.056°, and the image processing speed was more than 42 frames per second, which had high recognition accuracy and real-time performance. Based on the above, the online lane line recognition system based on the machine vision can effectively identify lane lines on roads in different environments and achieve the intervention of driving deviation, which can be applied to the lane keeping assist (LKA) system based on autonomous driving technology.
|
|
Received: 02 March 2020
Published: 28 August 2020
|
|
|
基于机器视觉的车道线在线识别系统设计
针对目前车道线识别系统因处理对象为离线图像或视频文件而难以实现车道线在线识别的问题，大多数车道线识别算法为了减少图像处理运行量，通常选取固定的兴趣区域（region of interest，ROI）进行处理，导致难以适应环境的动态变化且存在不同程度的识别误差。为此，提出了一种基于机器视觉的不设定兴趣区域的车道线在线识别系统。首先，利用VBAI（Vision Builder for Automation Inspection）平台对实时采集的彩色道路图像进行预处理，完成彩色图像的灰度化、滤波及二值化处理。然后，建立图像坐标系，并构建多条向外发散的灰度值采集线，灰度值采集线与车道线相交处的灰度值会发生较大改变，当某点的灰度值高于设定的灰度阈值时，记该点为边缘突变点。接着，借助直线拟合算法对所有车道边缘突变点进行拟合以完成车道线识别，并求解两侧内车道线的远方消失交点坐标和车辆行驶偏离车道中心线的相对航偏角，当相对航偏角超过不同等级的安全阈值时，系统人机交互界面的提示框呈现不同的颜色以进行提醒或预警。最后，借助LabVIEW进行API（application programming interface，应用程序接口）脚本调用，实现图像处理程序的连续运行与车道线在线识别。试验结果表明，所提出的车道线在线识别系统的识别准确率达98.41%以上，相对航偏角的测量误差小于0.056°，图像处理速度达42帧/s以上，兼具识别的准确性与实时性。综上可知，基于机器视觉的车道线在线识别系统可有效识别出不同环境路面的车道线，并实现行驶偏离预警，可将其应用于基于自动驾驶技术的车道保持辅助（lane keeping assist，LKA）系统。
|
Viewed |
|
|
|
Full text
|
|
|
|
|
Abstract
|
|
|
|
|
Cited |
|
|
|
|
| |
Shared |
|
|
|
|
| |
Discussed |
|
|
|
|
