In this project, I implemented a stereo camera-based visual odometry pipeline using the KITTI dataset to estimate vehicle pose through 3D motion tracking of surrounding features from both lenses. RANSAC-based outlier rejection filtered noisy feature correspondences, while scalar-weighted point cloud alignment and SVD yielded robust rotation and translation estimates. These transformations were accumulated over time to recover the vehicle’s trajectory. To address VO drift, the RANSAC framework ensured geometric consistency, limiting error accumulation. The system was fully developed in Python using OpenCV and NumPy, enabling self-contained state estimation without external localization.