Camera Relocalization Using Key Frames to Correct Planar Tracking Drift

The system pre-stores multiple key frames with their corresponding camera poses and feature points before tracking drifts too far. When relocalization is needed, the current frame is compared against these pre-stored key frames to find the best match, allowing the system to jump back to a known good pose without accumulating errors from continuous incremental tracking.

The system implements a feedback mechanism by continuously monitoring tracking quality metrics (such as feature match quality and pose consistency) and triggering relocalization when degradation is detected. This feedback loop allows the system to correct drift by returning to a previously verified accurate pose, thereby maintaining long-term tracking accuracy while preserving continuous tracking capability.

Instead of performing exhaustive feature matching with all stored key frames, the system uses partial matching strategies such as comparing against a subset of key frames, using approximate nearest neighbor search, or limiting the number of comparison iterations. This reduces computational overhead while still achieving sufficient accuracy for error correction.

Key frames are pre-processed and stored with extracted feature points and descriptors before they are needed for relocalization. This preliminary preparation allows the relocalization process to proceed more quickly by comparing against pre-computed data rather than processing raw images in real-time, thereby reducing the time penalty associated with frequent accuracy corrections.