Obstacle Detection Using Monocular Camera for Low Flying Unmanned Aerial Vehicle

This thesis describes the research of an obstacle detection system for a low flying autonomous unmanned aerial vehicle(UAV). The system utilized an extended Kalman filter based simultaneous localization and mapping algorithm which fuses navigation measurements with monocular image sequence to estimate the poses of the UAV and the positions of landmarks. To test the algorithm with real aerial data, a test flight was conducted to collect data by using a sensors loaded simulated unmanned aerial system(SUAS) towed by a helicopter. The results showed that the algorithm is capable of mapping landmarks ranging more than 1000 meters. Accuracy analysis also showed that SUAS localization and landmark mapping results generally agreed with the ground truth. To better understand the strength and weakness of the system, and to improve future designs, the algorithm was further analyzed through a series of simulations which simulates oscillatory motion of the UAV, error embedded in camera calibration result, and quantization error from image digitization.