摘要
文中主要论述了在无人机识别人物方面,为尽可能提高识别算法的实时性,提出了一种新的方案。首先对原始图像进行显著性区域提取使用全局对比度直方图(histogram-based contrast)方法,减少计算大部分不必要的背景区域。再对原图像显著性区域计算方向梯度直方图(Histogram of Oriented Gradient),使用非负矩阵分解(Non-negative Matrix Factorization,NMF)对已计算好的HOG降维,在保留了大部分原向量特征的条件下提高了鲁棒性和显著地降低了运算量。最后使用支持向量机(Support Vector Machine)对降维后的特征分类判断。相比与传统的HOG+SVM方法,本文在实时性方面有很大的提升,在嵌入式ARM平台运行达到27fps,基本达到实时识别跟踪的需求。
This paper mainly discusses the aspect of identifying the characters in the drone and proposes a new method to improve the real-time performance of the recognition algorithm.First,the saliency region extraction of the original image uses a global histogram-based contrast method to reduce the calculation of most unnecessary background regions.Then calculate the Histogram of Oriented Gradient for the original image saliency region,and use the Non-negative Matrix Factorization(NMF)to reduce the calculated HOG,and retain most of the original vector features.The robustness of the conditions is improved,and the amount of computation is significantly reduced.Finally,the support vector machine(Support Vector Machine)is used to classify the features after dimension reduction.Compared with the traditional HOG+SVM method,this paper has a great improvement in real-time performance.The embedded ARM platform runs at 27fps,which basically meets the requirements of real-time identification and tracking.
作者
彭文亮
梁祝
李智峰
PENG Wen-liang;LIANG Zhu;LI Zhi-feng(Zhuhai Campus Beijing Institute of Technology,Zhuhai 519088,China)
出处
《电子设计工程》
2019年第11期150-153,共4页
Electronic Design Engineering
