基于多特征融合的汽车先进辅助驾驶系统前方车辆检测方法

Front vehicle detection method in advanced driver assistance system based on multi-feature fusion

为提高先进驾驶辅助系统前方目标车辆检测的准确性,克服单一特征提取存在的漏检或检测结果不可靠缺点,提出了一种基于视觉的多特征融合的车辆检测方法。首先,建立了车辆尾灯对特征的约束条件并提取了图像中前方车辆的尾灯;接着,通过尾灯对质心距离确定了图像中车辆的宽度,从而获得了前方车辆标记区域;然后,根据提取的图像中路面像素灰度均值与标准差,确定车辆底阴影区域灰度值,实现了车底阴影区与路面背景灰度阈值的区分,进而确定了车底阴影区域,并结合车辆结构比例关系得到了目标车标记区;最后,针对阴影特征和尾灯对特征分别确定的标记区域进行整合处理,确定唯一的目标检测区域。针对提出的多特征融合的车辆检测方法,结合日间、夜间及阴雪等天气条件的大量图像,验证了方法的有效性。实验结果表明,所提出的基于多特征融合的汽车先进辅助驾驶系统前方车辆检测方法较单一特征检测法具有更广泛的适应性及更高的目标检测的可靠性。

In order to improve the accuracy of front vehicle detection in advanced driver assistance system and overcome the defects of single feature extraction,a vehicle detection method of multi-feature fusion was proposed based on vision.Firstly,the rear-lights in front of the vehicle were extracted by establishing the feature constraint condition of vehicle rearlights. Secondly,the vehicle width in the image was determined according to the distance between the centroid of the rearlights. Then,the gray threshold of the shadow region of the vehicle bottom and the road background were distinguished according to the average and standard deviation of the road pixel grayscale in the extracted image. Therefore,the gray value of the shadow region of the vehicle bottom was determined,and then the target vehicle marking areas was obtained according to the proportion relation of vehicle structure. Finally,a unique target detection area was identified by integrating the marking area determined by the shadow feature and rear-light feature respectively. The effectiveness of the proposed multifeature fusion vehicle detection method was verified on a large number of images of daytime,night,cloudy and snowy weather conditions. The experimental results show that,the proposed front vehicle method in advanced driver assistance system based on multi-feature fusion has wider adaptability and higher reliability compared with the single feature detection method.