Visual Odometry

Key Insight

Visual odometry (VO) estimates how a camera moved through the world using nothing but its own video, by tracking features — distinctive corners and textures — from frame to frame and back-solving the camera motion that explains how they shifted. Because each step's motion is measured relative to the previous frame, small errors compound into drift: travel 100 meters and your estimated path may be off by several, even though every single step looked accurate. This is why reporting drift over a fixed distance is the standard scorecard, and why a full SLAM system adds loop closure — recognizing a previously visited place — to snap the accumulated error back down.