融合两帧差分和背景差分的车辆检测算法  

Combined two-frame difference with background difference method for vehicle detection algorithm

作　　者：舒兆翰 李小龙[1,2,3] SHU Zhao-han;LI Xiao-long(Faculty of Geomatics,East China University of Technology,Nanchang 330013,China;Key Laboratory of Mine Environmental Monitoring and Improving around Poyang Lake of Ministry of Natural Resources,East China University of Technology,Nanchang 330013,China;CNNC Engineering Research Center of 3D Geographic Information,East China University of Technology,Nanchang 330013,China)

机构地区：[1]东华理工大学测绘工程学院,江西南昌330013 [2]东华理工大学自然资源部环鄱阳湖区域矿山环境监测与治理重点实验室,江西南昌330013 [3]东华理工大学中核三维地理信息工程技术研究中心,江西南昌330013

出　　处：《计算机工程与设计》2023年第4期1174-1180,共7页Computer Engineering and Design

基　　金：国家重点研发计划基金项目(2017YFB0503704);武汉大学测绘遥感信息工程国家重点实验室开放研究基金项目(20I01);江西省大气污染成因与控制重点实验室开放基金项目(AE2003)。

摘　　要：针对传统背景差分法和两帧差分法在车辆检测任务中出现的噪声、鬼影和空洞问题,提出一种将两者融合的静态背景下运动车辆检测算法。建立背景帧和检测帧的类别矩阵M1和M2,消除噪声和空洞;引入最大类方差法(Otsu)生成背景帧和检测帧的灰度级矩阵G1和G2,以此构造鬼影区的灰度级变化特征β,分割鬼影区域。实验结果表明,融合算法在测试集上的检测效果达到88.41%的平均F1值,相比传统的两帧差分、高斯混合模型和视觉背景提取算法均有大幅提升。Aiming at the problems of noise,ghost and holes in the vehicle detection task using the traditional background diffe-rence method and the two-frame difference method,a motion vehicle detection algorithm with a static background fusing these two methods was proposed.The category matrices M 1 and M 2 of background frames and detection frames were established to eliminate noise and holes.The Otsu method was introduced to generate the gray level matrices G 1 and G 2 of background frames and detection frames,and the gray level change featureβof the ghost region was constructed to segment the ghost region.Experimental results show that the fusion algorithm achieves an average F 1 value of 88.41%on the test set,which is a significant improvement over the traditional two-frame difference,Gaussian mixture model and visual background extraction algorithms.

关 键 词：运动车辆检测 静态背景 背景差分 两帧差分 类别矩阵 空洞 噪声 鬼影 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]