Research on pose estimation algorithm of non-cooperative target tracked vehicles based on PnP model

To address the pose estimation problem of non-cooperative tracked vehicles, this study proposes a non-iterative Perspective-n-Point (PnP) method. In particular, the method leverages the large contact area between the vehicle’s track and the ground, thereby ensuring that the pitch angle between the camera coordinate system and the object coordinate system remains largely consistent. Based on this constraint, the rotation matrix is simplified by fixing the pitch angle, which reduces the degrees of freedom in pose estimation. To further enhance robustness, the proposed framework employs the RANSAC algorithm to eliminate outliers, thereby reducing iterative errors associated with traditional PnP algorithms and filtering out mismatched feature points. Finally, the pose equation is solved by finding the local minimum of the cost function through algebraic optimization, thus avoiding convergence issues in iterative optimization. Experimental validation is conducted using an optoelectronic reconnaissance unmanned platform. The results demonstrate that the proposed approach not only simplifies the pose estimation process for tracked vehicles but also enhances computational efficiency. Notably, it achieves an improved computation rate while maintaining accuracy, thereby enabling real-time pose estimation. This advancement holds significant potential for applications in autonomous navigation and target tracking systems requiring rapid and reliable state estimation.