基于微波雷达回波信号的智能车道划分方法

发布时间：2018-05-16 10:46

  本文选题：多目标雷达 + 车道划分　； 参考：《计算机应用》2017年10期

【摘要】：利用多目标交通测速雷达进行交通执法时,只有正确地判断出车辆所在的车道,抓拍照片才能作为交通执法的依据。传统的分车道方法主要通过人工测量的固定阈值以及坐标系旋转的方法来达到车道划分的目的,但这种方法误差较大并且不易于操作。基于统计和密度特征的核聚类算法(K-CSDF)分两步进行:首先对雷达获取的车辆数据进行特征提取,包括基于统计特征的阈值处理和基于密度特征的动态半径提取;然后引入基于核的相似性的动态聚类算法对筛选出的有效点进行聚类。通过和高斯混合模型(GMM)算法以及自组织映射神经网络(SOM)算法进行仿真对比表明:当只取100个有效点进行聚类时,K-CSDF和SOM算法能达到90%以上的分车道正确率,而GMM算法不能给出车道中心线;在算法用时上,当取1000个有效点时,K-CSDF和GMM算法用时均小于1 s,可以保证实时性,而SOM算法则需要2.5 s左右;在算法鲁棒性上,K-CSDF对不均匀样本的适应性优于这两种算法。当取不同数量的有效点进行聚类时,K-CSDF可以达到95%以上的平均分车道正确率。
[Abstract]:When the multi-target traffic velocity radar is used to enforce traffic law, only by correctly judging the lane where the vehicle is located and taking pictures can it be used as the basis of traffic law enforcement. The traditional lane separation method mainly achieves the purpose of lane division by manual measurement of fixed threshold and rotation of coordinate system, but this method has a large error and is difficult to operate. The kernel clustering algorithm based on statistical and density features is divided into two steps: firstly, the vehicle data acquired by radar are extracted, including threshold processing based on statistical features and dynamic radius extraction based on density features; Then the kernel similarity based dynamic clustering algorithm is introduced to cluster the selected valid points. The simulation results with Gao Si hybrid model (GMM) algorithm and self-organizing mapping neural network (Som) algorithm show that the K-CSDF and SOM algorithms can achieve more than 90% accuracy of lane separation when only 100 effective points are taken for clustering. However, the GMM algorithm can not give the center line of the driveway, and when 1000 effective points are taken, the time of both K-CSDF and GMM algorithm is less than 1 s, which can guarantee the real time, while the SOM algorithm needs about 2.5 seconds. In terms of robustness, K-CSDF is better than these two algorithms in its adaptability to heterogeneous samples. When different number of effective points are used for clustering, K-CSDF can achieve an average accuracy of more than 95%.
【作者单位】： 北京信息科技大学通信工程系;北京川速微波科技有限公司;
【基金】：国家自然科学基金资助项目(61671069) 北京高等学校高水平人才交叉培养项目~~
【分类号】：TN958;U495
，

本文编号：1896526