Abstract:

A vehicle detection method with zero-height hypothesis validation was proposed to eliminate the disturbance caused by shadows and traffic signs on road surface. The principle was to assume that the initial detected targets were all 2D objects on road surface, the hypothesis was validated by the target movement during tracking process, and the targets which passed the validation were road surface disturbance. Feature-based method was applied on the image captured by monocular vision and the lidar data for the initial vehicle detection.The targets which failed to the hypothesis validation will be tracked, while the ones which passed the validation will be deleted. Experiments show that the proposed method can detect vehicles under various road conditions and can lower the false detection rate with the existence of road surface disturbance.

Key words: vehicle detection, zero-height hypothesis, monocular vision, intelligent vehicle

摘要：

提出了一种应用零高度假设检验的车辆检测方法以去除路面上阴影、交通标志等造成的干扰。该方法的原理是，假设初始检测所得到的所有目标物都是位于路面上的二维物体，利用目标物在跟踪过程中的位置变化对假设进行检验，通过假设检验的目标即为路面干扰。车辆的初始检测是利用基于特征的方法对单目视觉获取的图像和激光雷达获取的数据进行检测的，把未能通过假设检验的初始检测结果加入跟踪档案，删除那些通过假设检验的目标。试验表明，该方法能够在多种工况下对车辆进行识别，并能降低存在路面干扰情况下的识别误检率。

关键词: 车辆识别, 零高度假设, 单目视觉, 智能汽车