针对单目运动像机无法确定点状运动物体三维位置的难题,提出了一种求解点状运动物体三维位置的方法,利用直线在其法平面上的投影为点这个特性,将各时刻像机光心与物体位置的连线(以下称观察视线)投影到以运动物体轨迹为法线的平面上,通过迭代使其投影线交于一点,交点对应的直线就是物体的运动轨迹,仅需要取大于等于5个时刻的数据就可以确定物体的运动轨迹。假设物体作匀速直线运动,根据各时刻像机内外参数以及运动物体的成像列出线性方程组,得到物体运动轨迹的初值;将各时刻的观察视线投影到物体运动轨迹初值的法平面上,得到一系列投影视线的交点,并通过RANSAC方法剔除交点的野值;寻找这些RANSAC后的内点的最小外接圆,其圆心就是待求解的物体运动轨迹在法平面上的投影;通过几次迭代计算,就可以得到精度较高的定位结果。仿真和实验表明,该方法不仅对沿直线运动的物体有较好的定位精度,而且将一些复杂运动看作分段直线运动进行求解,也能够取得较好的结果。
In order to determine the 3D position of a moving point from a monocular moving camera, a positioning method is proposed by taking advantage of the property that the projection of a line on its normal plane is a point, the pro- jection lines of the line-of-sights on the plane which normal is the moving point trajectory intersect one point, and can get the trajectory of the point by 5 views. Firstly, that the moving point is constant velocity is supposed, the initial guess can be calculated by the linear equation set which need parameters of the camera and the images of the moving point. Second- ly, the line-of-sights are projected on the normal plane of the trajectory and some bad value can he eliminated by RANSAC from the intersection points of these the projection lines. The center of the smallest circle which satisfy all good intersection points is the projection of the moving point trajectory. The result with high precision can be obtained by sev- eral iterations. Simulation and experiment show that the method can get good results not only for the linear moving point but for the more complicated trajectory which can be divided into linear segments approximatively.