Highly Efficient Screening for Point-Like Targets via Concentric Shells

Details Highly Efficient Screening for Point-Like Targets via Concentric Shells Highly Efficient Screening for Point-Like Targets via Concentric Shells

Sep 2010

adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2010amos.confe..73w&link_type=abstract

Proceedings of the Advanced Maui Optical and Space Surveillance Technologies Conference, held in Wailea, Maui, Hawaii, September

Statistics

Computation

Scientific paper

We consider the problem of automatically screening for man-made objects (MMO) in infrared (IR) videos and synthetic aperture radar (SAR) imagery. Since hard targets are often highly reflective in SAR and also have an IR signature that differs from their surroundings, both problems reduce to finding point-like objects. Thresholding (usually locally adaptive) only utilizes the radiometric information and ignores the maximum target size, which means reflection artifacts or large regions are often returned as false alarms. Recently, a level-set approach has been proposed that takes speckle into account and reliably separates targets from the background [1]. However, its computational cost is almost certainly too high for large datasets or real-time video analysis. An alternative model called the "hotspot transform" was developed for IR Search and Track applications [2]. This operator (defined in Sec. 1) searches for local maxima that are entirely surrounded by a ring of darker pixels, thus suppressing bright but nonpoint- shaped regions. Its computational cost for N pixels and maximum target size R is O(N _ R _ R). We believe this technique to be suitable for screening in both IR and SAR data and have developed a novel algorithm that reduces its complexity to the optimal O(N _ R). Our sophisticated implementation, described in Sec. 2, avoids redundant computations with a divide and conquer scheme, ensures the working set fits in caches via pipelining, and achieves an additional 27-fold speedup via vectorization and parallelization. The attained processing rate of 72 MPixel/s on a single workstation enables screening entire satellite datasets within seconds (c.f. Sec. 3). Results are given for airborne SAR images and the MSTAR dataset in Sec. 4. The algorithm is shown to be suitable for detection of MMO and as a pre-processing step for multi-class target recognition via support vector machine (SVM).

Affiliated with

Also associated with

No associations

Rating

Highly Efficient Screening for Point-Like Targets via Concentric Shells does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.

If you have personal experience with Highly Efficient Screening for Point-Like Targets via Concentric Shells, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Highly Efficient Screening for Point-Like Targets via Concentric Shells will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFWR-SCP-O-1556011