结合Harris角点的栅格状雷达的检测方法  

An Approach to Detect Grid-like Radars Based on Harris Corner Recognition

作　　者：赵敏达[1] 李峰[1] 孙涛[2] 凌强[1] 

机构地区：[1]中国科学技术大学自动化系,合肥230022 [2]中国科学院长春光学精密机械与物理研究所光电对抗部,长春130033

出　　处：《小型微型计算机系统》2016年第5期1039-1043,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金面上项目(61273112)资助

摘　　要：栅格状雷达是现实中应用比较广泛的一种雷达,在复杂形变和场景下对其的检测具有重要的意义.阐述了一种新的栅格状雷达的检测方法,充分挖掘栅格状雷达的特征、提高了检测的鲁棒性和准确率.该方法首先对原图像进行一定的预处理以强化图像中的栅格形状特征,综合运用霍夫直线检测和Harris角点检测提取出直线交叉点等处的强角点,然后根据强角点的分布规律和密集程度,并结合基于颜色和位置的聚类结果,运用一定的判别准则判断图像中的图像中是否有雷达,并且在图像中有雷达的情况下将雷达区域分割出来.在由80张含有雷达和80张不含雷达的图片库上实验,在图像中有雷达时识别正确且分割区域正确的准确率可以达到88.75%,与经典算法"BOF+SIFT"和"HOG"相比拥有更好的性能;并且本方法针对亮度、尺度和旋转均具有不变性.As an important class of radars, Grid-like radars are widely applied in reality. So it makes great sense to detect grid-like ra- dars in complex deformation and scenes. This paper describes a new method for grid-like radar detection, It fully exploits the character- istics of a grid-like radar, thus improves the robustness and accuracy of the grid-like radar detection. It first performs pretreatments to enhance the grid-like characteristics of concerned radars. Then it implements the Hough transformation to detect straight lines and Harris comer detection method to extract strong Harris comers which lie on the detected straight lines. According to the distribution and density of Harris comers, together with the clustering results of pixels based on their color and positions ,it makes the decision whether a grid-like radar exists. Furthermore, it determines the region where the radar lies. Our method is tested with a dataset being composed of 80 grid-like radar images and 80 other images and its accuracy is 88.75%. The method was compared with traditional object detec- tion method：＂ BOF ＋ SIFT＂ ,＂ HOG＂. Our method demonstrates better robustness. Moreover, the approach is robust against illumination, scale and rotation.

关 键 词：K-MEANS聚类 角点检测 雷达检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]