Obstacle Detection for Low Flying Unmanned Aerial Vehicles Using Stereoscopic Imaging

This paper describes a stereoscopic imaging algorithm that is modified for obstacle detection in low flying unmanned aerial vehicles (UAVs). In this type of flight, obstacle detection must be carried out quickly for the system to be effective in real time. Additionally, since the aircraft is close to the ground, the horizon is usually at the top of the field of view and obstacles must be distinguished from the clutter of the terrain. The sparse edge detection and reconstruction algorithm proposed, produces fast but partial reconstructions of the environment. One image is passed through a series of edge detectors to generate a very sparse outline of the environment. This outline is then correlated to the second image and the resulting reconstruction is added to a model of the environment. Although each individual reconstruction is incomplete, the overall result after a short initialization period is a model of the environment that is more comprehensive than a single stereoscopic correlation run with a more detailed edge detector. Simulation of the algorithm on test image patterns showed an increase in performance relative to the length of the sequence of stereo pairs. On average, the signal to noise ratio (SNR) for sparse edge reconstruction was significantly higher than that for single correlation with more detailed edge detectors. Additionally it was found that the processing speed of the sparse edge detection algorithm on a pair of stereoscopic images is faster than the processing carried out by a more detailed edge detector. A test flight was also carried out to test the algorithm in a more realistic scenario. The test confirmed that the sparse edge reconstruction algorithm resulted in a much more detailed view of the environment than if a single, more detailed edge detector had been used.