针对目前运动目标检测领域内较流行的ViBe算法不适用于动态相机的问题,根据平面间变换为单应性变换的事实,提出了一种基于多重单应性变换的改进ViBe算法。利用ORB(oriented fast and rotated brief)特征点匹配与随机抽样一致(random sam...针对目前运动目标检测领域内较流行的ViBe算法不适用于动态相机的问题,根据平面间变换为单应性变换的事实,提出了一种基于多重单应性变换的改进ViBe算法。利用ORB(oriented fast and rotated brief)特征点匹配与随机抽样一致(random sample consensus,RANSAC)算法,计算得到相邻帧的不同平面的变换关系;采用全局光流补偿算法消除运动目标上的特征匹配点对,避免多重单应性变换中包含运动平面的变换;根据多重单应性变换改进ViBe算法的前景判定与背景模型更新,使改进算法适用于运动相机采集的数据。在公开视频数据集CDnet 2014与Complex Background Dataset上进行了相关实验,实验结果证明,改进算法能在尽量保留真实运动目标区域的同时,大幅度消除相机运动带来的影响,且在精度、特异性以及虚警率上表现出色,实时处理效率能达到20帧/s。展开更多
The authors study the asymptotic behavior of the smooth solutions to the Cauchy problems for two macroscopic models (hydrodynamic and drift-diffusion models) for semiconductors and the related relaxation limit problem...The authors study the asymptotic behavior of the smooth solutions to the Cauchy problems for two macroscopic models (hydrodynamic and drift-diffusion models) for semiconductors and the related relaxation limit problem. First, it is proved that the solutions to these two systems converge to the unique stationary solution time asymptotically without the smallness assump- tion on doping profile. Then, very sharp estimates on the smooth solutions, independent of the relaxation time, are obtained and used to establish the zero relaxation limit.展开更多
Let u=u（x,t,uo）represent the global solution of the initial value problem for the one-dimensional fluid dynamics equation ut-εuxxt＋δux＋γHuxx＋βuxxx＋f（u）x=αuxx,u（x,0）=uo（x）, whereα〉0,β〉0,γ〉0,δ〉0...Let u=u（x,t,uo）represent the global solution of the initial value problem for the one-dimensional fluid dynamics equation ut-εuxxt＋δux＋γHuxx＋βuxxx＋f（u）x=αuxx,u（x,0）=uo（x）, whereα〉0,β〉0,γ〉0,δ〉0 andε〉0 are constants.This equation may be viewed as a one-dimensional reduction of n-dimensional incompressible Navier-Stokes equations. The nonlinear function satisfies the conditions f（0）=0,｜f（u）｜→∞as ｜u｜→∞,and f∈C^1（R）,and there exist the following limits Lo=lim sup/u→o f（u）/u^3 and L∞=lim sup/u→∞ f（u）/u^5 Suppose that the initial function u0∈L^I（R）∩H^2（R）.By using energy estimates,Fourier transform,Plancherel＇s identity,upper limit estimate,lower limit estimate and the results of the linear problem vt-εv（xxt）＋δvx＋γHv（xx）＋βv（xxx）=αv（xx）,v（x,0）=vo（x）, the author justifies the following limits（with sharp rates of decay） lim t→∞[（1＋t）^（m＋1/2）∫｜uxm（x,t）｜^2dx]=1/2π（π/2α）^（1/2）m!!/（4α）^m[∫R uo（x）dx]^2, if∫R uo（x）dx≠0, where 0!!=1,1!!=1 and m!!=1·3…（2m-3）…（2m-1）.Moreover lim t→∞[（1＋t）^（m＋3/2）∫R｜uxm（x,t）｜^2dx]=1/2π（x/2α）^（1/2）（m＋1）!!/（4α）^（m＋1）[∫Rρo（x）dx]^2, if the initial function uo（x）=ρo′（x）,for some functionρo∈C^1（R）∩L^1（R）and∫Rρo（x）dx≠0.展开更多
文摘针对目前运动目标检测领域内较流行的ViBe算法不适用于动态相机的问题,根据平面间变换为单应性变换的事实,提出了一种基于多重单应性变换的改进ViBe算法。利用ORB(oriented fast and rotated brief)特征点匹配与随机抽样一致(random sample consensus,RANSAC)算法,计算得到相邻帧的不同平面的变换关系;采用全局光流补偿算法消除运动目标上的特征匹配点对,避免多重单应性变换中包含运动平面的变换;根据多重单应性变换改进ViBe算法的前景判定与背景模型更新,使改进算法适用于运动相机采集的数据。在公开视频数据集CDnet 2014与Complex Background Dataset上进行了相关实验,实验结果证明,改进算法能在尽量保留真实运动目标区域的同时,大幅度消除相机运动带来的影响,且在精度、特异性以及虚警率上表现出色,实时处理效率能达到20帧/s。
基金Project supported by the National Natural Science Foundation of China, the Grant of MST of China,the National Natural Science
文摘The authors study the asymptotic behavior of the smooth solutions to the Cauchy problems for two macroscopic models (hydrodynamic and drift-diffusion models) for semiconductors and the related relaxation limit problem. First, it is proved that the solutions to these two systems converge to the unique stationary solution time asymptotically without the smallness assump- tion on doping profile. Then, very sharp estimates on the smooth solutions, independent of the relaxation time, are obtained and used to establish the zero relaxation limit.
文摘Let u=u（x,t,uo）represent the global solution of the initial value problem for the one-dimensional fluid dynamics equation ut-εuxxt＋δux＋γHuxx＋βuxxx＋f（u）x=αuxx,u（x,0）=uo（x）, whereα〉0,β〉0,γ〉0,δ〉0 andε〉0 are constants.This equation may be viewed as a one-dimensional reduction of n-dimensional incompressible Navier-Stokes equations. The nonlinear function satisfies the conditions f（0）=0,｜f（u）｜→∞as ｜u｜→∞,and f∈C^1（R）,and there exist the following limits Lo=lim sup/u→o f（u）/u^3 and L∞=lim sup/u→∞ f（u）/u^5 Suppose that the initial function u0∈L^I（R）∩H^2（R）.By using energy estimates,Fourier transform,Plancherel＇s identity,upper limit estimate,lower limit estimate and the results of the linear problem vt-εv（xxt）＋δvx＋γHv（xx）＋βv（xxx）=αv（xx）,v（x,0）=vo（x）, the author justifies the following limits（with sharp rates of decay） lim t→∞[（1＋t）^（m＋1/2）∫｜uxm（x,t）｜^2dx]=1/2π（π/2α）^（1/2）m!!/（4α）^m[∫R uo（x）dx]^2, if∫R uo（x）dx≠0, where 0!!=1,1!!=1 and m!!=1·3…（2m-3）…（2m-1）.Moreover lim t→∞[（1＋t）^（m＋3/2）∫R｜uxm（x,t）｜^2dx]=1/2π（x/2α）^（1/2）（m＋1）!!/（4α）^（m＋1）[∫Rρo（x）dx]^2, if the initial function uo（x）=ρo′（x）,for some functionρo∈C^1（R）∩L^1（R）and∫Rρo（x）dx≠0.
