论文部分内容阅读
提出一种融合多种目标特征的单目视觉车辆检测方法。首先,利用车辆尾部的结构对称性提取出感兴趣区域(ROI),减少搜索范围;然后,利用车辆底部的阴影特征,在ROI中搜寻车辆可能出现的位置,找出假设目标;最后,利用亮度和轮廓信息对假设目标进行对称性验证,排除虚假目标,同时对车辆在图像中的位置实现精确定位。通过实验,验证了提出方法的有效性和鲁棒性。
A monocular vision vehicle detection method combining multiple target features is proposed. Firstly, the ROI is extracted by using the structure symmetry of the tail of the vehicle to reduce the search range. Then, the shadow feature at the bottom of the vehicle is used to search the ROI for the possible location of the vehicle to find the hypothetical target. Finally, And contour information of the hypothetical target symmetry verification, eliminate false targets, while the location of the vehicle in the image to achieve precise positioning. Experiments show that the proposed method is effective and robust.