GPU processing for parallel image processing and real-time object recognition

被引：0

作者：

Vincent, Kevin ^{[1
]}

Damien Nguyen ^{[3
]}

Walker, Brian ^{[4
]}

Lu, Thomas ^{[2
]}

Chao, Tien-Hsin ^{[2
]}

机构：

[1] Calif State Univ Fullerton, Fullerton, CA 92634 USA

[2] CALTECH, Jet Prop Lab, Pasadena, CA 91125 USA

[3] Saddleback Coll, Mission Viejo, CA 92692 USA

[4] Georgia Inst Technol, Atlanta, GA USA

来源：

OPTICAL PATTERN RECOGNITION XXV | 2014年 / 9094卷

基金：

美国国家航空航天局;

关键词：

automated target recognition; GPU; CUDA programming; parallelization; real-time; computer vision; optimization;

D O I：

10.1117/12.2054353

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

In this paper, we present a method for reducing the computation time of Automated Target Recognition (ATR) algorithms through the utilization of the parallel computation on Graphics Processing Units (GPUs). A selected multi-stage ATR algorithm is refounded to encourage efficient execution on the GPU. Such refounding includes parallel reimplementations of optical correlation, Feature Extraction, Classification and Correlation using NVIDIA's CUDA programming model. This method is shown to significantly reduce computation time of the selected ATR algorithms allowing the potential for further complexity and real-time applications.

引用

页数：12