In this paper, we propose an optimized trajectory recovery algorithm for on-road vehicles from monocular videos, which considers multiple constraints between frames in a sliding window. We build a linear system to optimally estimate the camera rotation matrix corresponding to the first frame in each sliding window, and calculate the camera center estimation. The camera centers are determined by the estimation and prediction eventually. The whole process is applied to the sequential video frames by sliding the window. The advantage of the proposed algorithm is that it is able to reduce the drift error, which is a challenge for the existing algorithms, since multiple constraints in a sliding window are used to optimize the camera poses. We test the algorithm both on synthetic data and real data, and compare it with existing algorithms. It shows a higher accuracy of reconstruction and is suitable to recover a better trajectory of on-road vehicles from monocular videos.