DragonflEYE: a passive approach to aerial collision sensing

"This dissertation describes the design, development and test of a passive wide-field optical aircraft collision sensing instrument titled 'DragonflEYE'. Such a ""sense-and-avoid"" instrument is desired for autonomous unmanned aerial systems operating in civilian airspace. The instrument was configured as a network of smart camera nodes and implemented using commercial, off-the-shelf components. An end-to-end imaging train model was developed and important figures of merit were derived. Transfer functions arising from intermediate mediums were discussed and their impact assessed. Multiple prototypes were developed. The expected performance of the instrument was iteratively evaluated on the prototypes, beginning with modeling activities followed by laboratory tests, ground tests and flight tests. A prototype was mounted on a Bell 205 helicopter for flight tests, with a Bell 206 helicopter acting as the target. Raw imagery was recorded alongside ancillary aircraft data, and stored for the offline assessment of performance. The ""range at first detection"" (R0), is presented as a robust measure of sensor performance, based on a suitably defined signal-to-noise ratio. The analysis treats target radiance fluctuations, ground clutter, atmospheric effects, platform motion and random noise elements. Under the measurement conditions, R0 exceeded flight crew acquisition ranges. Secondary figures of merit are also discussed, including time to impact, target size and growth, and the impact of resolution on detection range. The hardware was structured to facilitate a real-time hierarchical image-processing pipeline, with selected image processing techniques introduced. In particular, the height of an observed event above the horizon compensates for angular motion of the helicopter platform."