Oleg Gusyatin (MIT Lincoln Laboratory)
Keywords: image processing, segmentation, optical detection, space situational awareness
Abstract:
Depending on the mode of operation, an optical sensor records signatures of sidereal and non-sidereal objects as points or streaks. Streaks that result from fast moving objects present a challenge to automated detection algorithms since potential for heterogeneity in the expected optical signature is higher than that of a point-like object due to the fact that energy is spread over a larger number of pixels during integration. Such heterogeneities may arise from object tumbling, atmospheric effects, occlusions, or even focal plane irregularities. While standard image processing techniques rely on a single estimated threshold applied to a field of view to isolate sidereal and non-sidereal pixels from background it is often the case that objects captured as longer streaks are being fragmented into multiple collections of pixels resulting in inaccurate position predictions. Such erroneous position estimates often cause loss of detection. Here we present a novel segmentation algorithm that maximizes fidelity of the resulting streaking objects on a focal plane for highly heterogeneous signatures. Our technique is general in that it assumes that any given objects signature is separable from the background by multiple thresholds simultaneously allowing for accurate segmentations in high-noise or even occluded settings. Segmentation algorithm that we propose is based on a dynamic region-growing technique where a decision for including each individual pixel into a given object is made based on both statistical and spatial properties of the object. Such decisions are made dynamically as objects are being segmented. This segmentation algorithm is robust and its complexity order does not exceed that of standard segmentation techniques, making it an attractive alternative to signal enhancement techniques such as matched filtering. Testing has been conducted on the ground-based acquisitions of Low-Earth Orbit (LEO) satellites where, in certain datasets, number of detected streaking stars has been increased by over 20%. Also, we have observed a significant increase in detection rates of streaking satellites in high-fidelity simulation data from the Space-Based Space Surveillance program.
Date of Conference: September 14-17, 2010
Track: Posters