针对前视红外图像中地面固定目标的识别问题,提出了一种基于三维模型的匹配识别方法.首先由场景的3D数据建立目标三维模型,并以人工标记的方式进行编号以保留交界线信息;然后根据实时观测参数进行二维投影绘制得到目标的二维模板图像;最后提取边缘加权HOG特征在观测图像中进行匹配.对大量实测数据的实验结果表明,该方法识别精度高、对噪声、遮挡与参数偏差的鲁棒性好,在各种视角、距离、环境条件以及复杂背景的情况下均能达到很好的识别性能.该方法具有很强的实用价值,同时对于精确制导、自主导航等领域的研究也提供了新的思路和途径.%Aimed at ground target recognition in forward looking infrared(FLIR) images,a novel method based on 3D model was proposed. Firstly, a 3 D model of the target is constructed from the 3 D data of the scene, and is manually labeled to reserve the borderlines. Then 2D template images are obtained by projecting the 3D model with the real-time observation parameters. Finally, edge weighted HOG features are extracted and matched. The experiment result from plenty of real data proves that the proposed method is effective and robust, and performs very well in any case of view point, distance, environment and complex background. This method has great practical value, while it also provides new ideas and approaches for the precision guidance, autonomous navigation and other research areas.
展开▼
