嵌入式平台下红外图像坦克目标识别的实现方法

Realization of Target Recognition of Infrared Image Tank Based on Embedded Platform

针对战场环境下红外图像中坦克目标在嵌入式系统中的快速识别,提出一种改进的HOG特征提取结合AdaBoost分类算法的红外图像坦克目标识别实现方法。为了便于在ARM/FPGA嵌入式平台实现,该方法根据FPGA的计算特性对HOG特征提取算法进行优化设计,克服FPGA不适合的除法运算与反正切运算带来弊端。为验证算法的有效性,通过软硬协同设计的方法搭建了嵌入式系统,开发了ARM/FPGA样机,采集坦克目标的红外样本数据,在不同平台开展了目标识别实验。实验结果表明:改进后的算法在保证识别率的基础上,单张图片的处理时间平均为52ms,大幅地提高了运算速度。

Aiming at the problem of fast recognition of tank target in infrared image in embedded system in battlefield environment,an improved hog feature extraction method combined with AdaBoost classification algorithm was proposed.To be applied in the architecture of embedded system for ARM/FPGA controller,this method optimized the feature extraction of HOG according to the calculation characteristics of FPGA.After optimization,it avoids the division operation and arctangent operation that would consume time for FPGA.In order to verify the effectiveness of the algorithm,the embedded system is setup in software and hardware collaborative architecture.The prototype of ARM/FPGA controller was also developed for it.Testing experiments were carried out in different operating platforms.The experimental results show that the average processing time of single image is about 52 ms,which could dramatically shorten the real running time.