基于视图不确定关系模型的地面目标位置估计和跟踪

Position Estimation and Tracking Based on View Uncertainty Relational Model for Targets on A Ground Plane

为了实现地面上目标的位置估计和跟踪,提出了一种摄像机平面视图与地面位置估计之间关系的建模理论,推导了正态随机变量在投影下的变换;首先分析了随机变量(摄像机图像平面上的位置估计和地面上的位置估计)在投影变换下的变换方式,将变换后的随机变量中的Cauchy分量的强度与图像平面上目标的真实位置到地面无穷远处的直线投影之间的距离联系起来,表明了当某些几何性质满足时,投影变换会将正态分布映射为正态分布;其次采用无迹变换来计算得到变换后的随机变量的矩;最后采用得到的建模相关性设计了一种用于多个摄像机位置估计的最小方差估计器,并应用于跟踪地面上动态环境中的多个目标;实验结果表明,提出的模型不仅具有较好的组合位置估计能力,而且还能够利用这种模型得到的最小方差估计器有效地呈现和跟踪地面目标。

In order to achieve the position estimation and tracking of targets on a ground plane.a theory for modeling the relation between the camera plane view and position estimation on a ground plane is proposed,and the transformation of normal random variables under projection is deduced.Firstly,the transformation mode of random variables(position estimation on camera image plane and position estimation on a ground plane)under a projective transformation is analyzed,the strength of Cauchy component in the transformed random variable is associated with the distance of the true location of the target in the image plane from the projection of the line at infinity of the ground plane,which shows that when some geometric properties are satisfied,a projective transformation maps normal distributions to normal distributions.Secondly,the unscented transformation is used to calculate the moments of the transformed random variables.Finally,the obtained modeling correlation is used to design a minimum variance estimator of the position estimation by multiple cameras,and the tracking of multiple targets is applied in a dynamic environment on a ground plane.The experimental results show that the proposed model not only has a good combined position estimation ability,but also can effectively present and track targets on a ground plane by using the obtained minimum variance estimator of the model.