基于轨迹稀疏聚类的高速公路车辆检测  

Highway Vehicle Detection Based on Sparse Trajectory Clustering

作　　者：杨露 宋焕生[1] 张朝阳[1] YANG Lu;SONG Huansheng;ZHANG Zhaoyang(School of Information Engineering,Chang’an University,Xi’an 710064,China)

机构地区：[1]长安大学信息工程学院,西安710064

出　　处：《计算机工程与应用》2020年第15期251-258,共8页Computer Engineering and Applications

基　　金：教育部联合基金(No.6141A02022610);国家自然科学基金面上项目(No.61572083);陕西省重点研发计划重点项目(No.2018ZDXM-GY-047)。

摘　　要：针对高速公路中车辆的实时检测问题,提出了一种基于轨迹稀疏谱聚类的高速公路车辆检测方法。使用ORB算法检测特征点并利用基于金字塔LK光流算法进行跟踪得出特征点轨迹,将轨迹逆投影至三维世界坐标系,利用轨迹三维信息构建轨迹间的相似矩阵并对其进行稀疏化处理,采用谱聚类方法对特征点轨迹进行初步聚类,对谱聚类结果进行类间合并得出车辆检测结果。实验结果表明,方法花费了更少的时间代价,有效地解决了车辆遮挡问题,车辆实时检测精度提高至93%,具有一定的有效性和价值。To solve the real-time vehicle detection problem in expressway,a vehicle detection method based on trajectory sparse spectral clustering is proposed.The main processing adopts ORB algorithm to detect feature points,tracks the trajectory of these feature points by Pyramid LK optical flow algorithm.The next step is inversely projecting trajectory to 3D world coordinate system.At the same time,similarity matrix is constructed with 3D trajectories information and is processed sparsely.After preliminary clustering the feature points,vehicle detection result is obtained by merging classes.Experimental results show that proposed method effectively solves vehicle occlusion problem and vehicle detection accuracy is improved to 93%,which has certain validity and value.

关 键 词：车辆检测 轨迹聚类 谱聚类 稀疏化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]