单目序列的刚体目标位姿测量

针对单目相机不具备构建合作目标的情况及其对平面上运动目标位姿测量的应用需求，提出一种基于目标轮廓的运动目标位姿自动解算方法.该方法以目标三维几何模型和相机参数作为先验信息，利用OpenGL生成位于不同位姿的运动目标模拟图像，通过模拟图像与实拍图像之间的关系进行位姿的求解.首先由离散位姿集合的模拟图像得到运动目标轮廓集合，由实拍图像运动目标轮廓与离散位姿轮廓集合中轮廓间的关系确定运动目标的初始位姿.然后构建初始位姿下模拟图像的目标轮廓与实拍图像目标轮廓间的距离代价函数，采用非线性优化算法迭代求解运动目标的精确位姿.实验结果表明，提出的方法可以有效地测量运动目标的位姿，且当图像中纹理复杂、受阴影影响时仍能取得较好的测量结果.
Abstract: An automatic method for the position and orientation achievement of moving object based on contour was proposed, since the monocular camera cannot construct the cooperative target and the position and orientation measurement to the moving object on the plane is required. The proposed method was under the prior condition of three-dimensional geometric model of the rigid object and the camera parameters. OpenGL technology was used to generate the simulated images of the moving objects located in different positions and orientations. The position and orientation were solved by the relationship between the simulated images and the real images. Firstly, the moving object contour set was obtained according to the simulation of discrete position and orientation sets, and the initial position and orientation of the moving object was determined from the relationship between the moving object contour of real image and the contour in the discrete position and orientation contour set. Then, the distance cost function between the object contour of the simulated image of initial value and the object contour of the real image was constructed, and the precise position and orientation of the moving object was solved by iterative nonlinear optimization algorithm. The experimental results show that the proposed method can effectively measure the position and orientation of the moving object, and obtain reliable measurement results of the image with complex texture and shadow.