摘要
激光成像雷达能成清晰的目标三维距离像和一维强度像,可提高目标识别率,因而成为国际上的研究热点。当大视场高分辨率激光成像雷达垂直探测目标时,视场内目标增多,要求目标识别算法既能同时检测多目标,又要具有平面内旋转不变性。为了满足上述要求,提出将具有平面内旋转不变性的CHF-MACH相关滤波器和支持向量机(SVM)相结合,组成一种新的目标识别系统,其中相关滤波器能同时检测定位多个感兴趣目标,再用SVM分别对图像内的已定位的目标进行识别。以仿真激光成像雷达图像为实验数据,分别对4类目标进行识别。实验结果表明,CHF-MACH滤波器对本类目标有较好的检测率,对非本类目标有一定的抑制作用;SVM能以较高的精度分类已检测目标。所以,该方法能有效地对大视场内多目标进行识别,适用于激光成像雷达。
Laser radar can output the clear 3D range image and 1D intensity image, and it can improve the target recognition accuracy. So it becomes the international hot topic. When the laser radar with wide field of view （FOV） and high resolution vertically detects the targets, there are multiple targets in FOV. Therefore, the target recognition not only needs to detect the multiple targets, but also requires the in-plane rotation invariance. In order to meet the above requirements, a new target recognition system was proposed, which combined the CHF-MACH filter with the in-plane rotation invariance and SVM. The filter could simultaneously detect and locate the multiple interesting targets, and then the SVM was used to recognize the located targets in the image. Using the simulation images as the experiment data, four kind of targets were recognized. The results show that the CHF-MACH filter has the well detection performance for the similar class targets, and can restrain the other class targets. SVM can classify the detected targets with high precision. So, the method can effectively recognize the multiple targets in FOV, and it is suitable for the laser radar.
出处
《红外与激光工程》
EI
CSCD
北大核心
2008年第5期916-919,共4页
Infrared and Laser Engineering
基金
哈尔滨工业大学优秀团队支持计划资助项目
关键词
目标识别
激光成像雷达
支持向量机
相关探测
Target recognition
Imaging laser radar
Support vector machine
Correlation detection
