基于多传感器的地面移动目标识别技术研究

摘要

Based on traditional ground target recognition systems, a new ground target recognition system based on multi-sensor information fusion, using visible light sensors and infrared sensors, is developed in order to get effective target information in the cases of night, rain, snow and smog. Firstly, Gaussian template mean filter is adopted to the visible light image, and median filter is adopted to the infrared image. And then, the inter frame differ⁃ence method and background difference method are combined to realize the detection of the moving target. An algo⁃rithm based on the coordinates of the center of gravity connected domain mark is presented and the target area is lo⁃cated successfully. Secondly, the Haar-like rectangular features are used to represent the vehicle, the original char⁃acteristics library is expanded and rotation single rectangle features are added to describe the vehicle shadow re⁃gion. Finally, the traditional AdaBoost algorithm is improved and the visible light and infrared vehicle classifier is trained successfully. Experimental results show that the system has some practical value and research significance.%在传统的地面目标识别系统上，综合运用可见光和红外传感器，开发了基于多传感器信息融合的地面目标识别系统，有效地解决了在夜晚、雨雪天气和浓烟浓雾遮挡的情况下，能够获取有效的目标信息。首先，分析研究确定对可见光图像进行高斯模板均值滤波，对红外图像进行中值滤波处理。然后运动目标检测运用了帧间差分法和背景差分法结合的检测算法。并且研究了一种基于连通域重心坐标标记的算法，并成功定位了目标区域。其次，使用Haar-like矩形特征来表达车辆，对原有特征库进行扩展，添加旋转单一矩形特征来描述车底阴影区域。最后，对传统AdaBoost算法作出了改进，成功训练出可见光和红外车辆分类器。结果证明，该系统具有一定实用价值和研究意义。