An adaptive cubature Kalman filter algorithm based on singular value decomposition (SVD-ACKF) was proposed for orbit determination by optoelectronic theodolite when unknown or inaccurate noise statistics lead to low precision and divergence of filter. First, Sage-Husa maximum a posteriori and its improved form were used to estimate noise statistics online, and SVD instead of Cholesky decomposition in was used order to improve the stability of numerical calculation. Then, the mathematical model of orbit determination was expound, compared with the Euler method, improved Euler method was used to disperse the orbital dynamics equations with J2 perturbation. Finally, the simulation results show that improved Eular method can achieve a higher discrete precision, and SVD-ACKF algorithm can improve the accuracy and stability.