基于视觉EPnP加权迭代算法的三维位移实时测量

Vision based real-time 3D displacement measurement using weighted iterative EPnP algorithm

单目视觉三维位移测量的关键在于获取相机位姿参数,该问题可通过n点透视(PnP)算法求解。为提高PnP算法精度,提出一种改进的EPnP加权迭代算法(WIEPnP)。WIEPnP通过对标志点设置权重系数,再进行迭代计算,从而降低标志点深度和图像噪声对算法性能的影响。用MATLAB仿真实验对比研究了6种PnP改进算法,结果表明,WIEPnP算法能有效降低标志点深度的影响并有效降低图像高斯噪声对算法结果的影响,且算法精度和耗时均满足现场应用需求。WIEPnP的有效性同样在样机实验中得到了验证:样机在x、y方向的测量误差均小于1 mm;在z方向,WIEPnP算法有效降低了深度变化的影响,使z方向的绝对误差也不大于3 mm。可见提出的WIEPnP算法在实时性和误差性能方面都具有较好性能,能够满足大多数三维位移实时测量要求。

The key of monocular vision based 3D displacement measurement is to obtain the camera pose parameters, which can be achieved by solving a perspective n points(PnP) problem. In order to improve the accuracy of PnP algorithm, this paper proposes an improved weighted Iterative EPnP algorithm(WIEPnP). WIEPnP intends to reduce the influence from sign point depth and image noise on the algorithm performance. It is done by setting weight coefficients for sign points and then conducting iterative calculations. MATLAB simulation experiments were carried out for comparative study with 6 PnP algorithms. The results show that the newly proposed WIEPnP can effectively reduce the impacts from the sign point depth and the effect of image Gaussian noise, respectively;and its accuracy and computation time satisfy the field application requirements. Later, the prototype experiments also verify the effectiveness of WIEPnP. In the prototype experiments, measurement errors in x and y directions are convinced to be less than 1 mm. In terms of z direction, the WIEPnP algorithm can effectively reduce the effect of depth changes;thus, the absolute error in z direction is restricted to no more than 3 mm. It can be seen that the WIEPnP algorithm proposed in this paper has good performance in terms of real-time and error;and it can meet the requirements for most real-time 3D displacement measurement.