构建多部件关系概率模型的车辆检测方法

发布时间：2018-06-12 11:13

本文选题：车辆检测 + 多部件模型　；参考：《西安电子科技大学学报》2017年03期

【摘要】：传统车辆检测方法在处理交通复杂场景时,由于判断依据单一而容易出现车辆被误检和漏检的情况.为此,笔者提出一种基于车尾灯对和车牌之间空间位置关系概率模型的车辆检测方法.首先对道路交通视频图像进行颜色模型转换,分别得到突显车牌区域和尾灯区域的灰度图像,然后,对车牌灰度图像,通过梯度特征提取、区域梯度平滑和局部极大梯度查找几个关键步骤完成车牌的进一步准确定位,同时,对尾灯灰度图像经过阈值分割、连通域分析和面积计算几个关键步骤完成尾灯对的进一步准确定位.最后,建立尾灯和车牌在同一空域下的位置关系概率模型,并计算出该模型的概率似然度,进而判断部件所属的目标是否是同一辆车.实验结果表明,引入部件的空间位置关系概率模型,可以明显提高车辆检测的精度,且对不同型号的车辆检测鲁棒性都较好.
[Abstract]:When the traditional vehicle detection method is used to deal with the complicated traffic scene, it is easy to misdetect and miss the vehicle because of the single judgment basis. Therefore, a vehicle detection method based on the probabilistic model of the spatial position relationship between the taillight pair and the license plate is proposed. Firstly, the color model of the road traffic video image is transformed to get the gray image of the license plate area and the taillight area respectively. Then, the gray image of the license plate is extracted by gradient feature. Several key steps of region gradient smoothing and local maximum gradient searching are used to locate the license plate accurately. At the same time, the image of taillight grayscale is segmented by threshold. Several key steps of the connected domain analysis and area calculation are to complete the further accurate location of the taillight pair. Finally, the probability model of the position relationship between taillight and license plate in the same airspace is established, and the probability likelihood degree of the model is calculated to determine whether the target of the component belongs to the same vehicle. The experimental results show that the accuracy of vehicle detection can be improved obviously by introducing the probability model of spatial position relationship of components, and the robustness of vehicle detection is good for different types of vehicles.
【作者单位】：长安大学信息工程学院;西藏民族大学信息工程学院;北京交通大学交通运输学院;
【基金】：国家自然科学基金资助项目(61572083) 西藏科技厅自然科学基金资助项目(2015ZR-13-17)
【分类号】：TP391.41

【相似文献】