摘要
为了解决汽车安全辅助驾驶系统中的前向车辆检测问题,提出了一种基于单目视觉的在线前向车辆检测系统。通过检测车底阴影特征来生成车辆假设,分别提出了自适应路面阈值方法和阴影区域融合方法以解决路面区域灰度变化和阴影边缘变形问题;使用基于梯度特征的adaboost方法来验证车辆假设;最后使用Kalman滤波对检测到的目标进行跟踪以改善系统性能。使用道路实拍的图像序列对系统进行了测试。结果表明,该系统能够在实时条件下有效检测前方车辆。
A single camera based on road forward vehicle detection system was developed to detect vehicles in front of the driver in a driver assistance system to improve vehicle safety. The shadows underneath vehicles are detected to generate vehicle hypotheses. An adaptive road threshold method and a dark areas merge approach are used to resolve the grayscale variation on the road and the distortion of the shadow border. A gradient based adaboost method is used to verify the hypotheses. Kalman filters are then used to track objects to improve system performance. The system was tested with series of images taken on roads. The results show that the system can effectively detect vehicles in front of the driver on real time.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2007年第10期1713-1716,共4页
Journal of Tsinghua University(Science and Technology)
基金
清华-OMRON合作项目(0506J52)
