摘要
路面检测在基于视觉导航的汽车自动驾驶中具有重要意义。针对路面的多样性和场景的复杂性,文中研究并开发了基于DSP的路面检测技术,提出了一种融合图像灰度和图像深度的路面检测方法。系统首先采用双目立体视觉方法获取场景深度图像,然后根据图像灰度和图像深度进行超像素分割,最后提取超像素的灰度和深度特征,用SVM分类器对超像素进行分类,实现路面的检测。实际场景的测试证明了文中方法的可行性。街景图像数据库的实验表明,文中路面检测方法的准确性高。
Road detection is important in vision-based navigation in vehicle automatic driving. For the diversity of road and the complexity of the scene,this paper studied and developed road detection technique based on DSP,and proposed a road detection method,which fused the image intensity and the image depth. The system first used binocular stereo vision method to get the scene depth image,and then the image was segmented into super pixels according to the image intensity and the image depth,finally the road detection was realized through superpixels' gray and depth features,and superpixels were classified with SVM realizing. The actual scenario testing verifies the feasibility of the method. Experiments on the street image database show that the road detection method in this paper has high accuracy.
出处
《仪表技术与传感器》
CSCD
北大核心
2015年第12期110-112,共3页
Instrument Technique and Sensor