Vision Based Navigation System For An Aircraft in Approach Phase

This paper presents an event driven linear Kalman filter that fuses asynchronous vision based camera pose estimates with simulated onboard IMU and radio altimeter data to improve position, velocity, and altitude estimation during approach. Implemented in X Plane for manual 3 nm and 10 nm approaches, the filter smooths noisy vision outputs, rejects spurious peaks, maintains continuity through vision dropouts, and closely tracks radio altimeter altitude with final approach errors under ~30 ft. Post processed vision attitude estimates follow the recorded data from X-Plane flight simulator with larger outliers at longer ranges. Results support using this fusion approach for autonomous landing guidance and motivate angular rate fusion and Machine Learning based pose estimation next.