基于可见光图像的近红外场景仿真

Near Infrared Scene Simulation Based on Visual Image

利用地物光谱仪采集了晴天条件下典型地物在380-1100 nm的光谱反射率,并利用经过定标后的CMOS相机采集典型地物不同时段的可见光、近红外图像,结合相机辐射定标结果和地物目标的反射率建立了可见光、近红外图像灰度值的映射关系,且对简单场景进行图像分割,并通过查找表快速将可见光图像转化为近红外图像。建立了太阳辐射模型,在反演精度要求不是很高的条件下,可以反演一天内任意时刻的近红外图像。仿真结果表明,本文方法可生成真实感较好的近红外图像,为后续的各时段、各种天气条件及多种地物的近红外场景仿真奠定基础。

In this paper we used a spectrometer to collect the spectral reflectance of typical ground objects at 380-1100 nm under sunny conditions and a calibrated CMOS camera to collect images of the typical objects in visible and near infrared wavelengths at different time of the day. We established a gray value mapping relationship between visible images and near-infrared images, combining camera radiation calibration results and different reflection characteristics of ground objects. Also we exerted segmentation on simple scenes and converted visible images to near infrared images by the above established mapping relationship. In addition, we built a solar radiation model. We could inverse near infrared images at any time in a day, if the requirement of inversion accuracy is not very high. The simulation results show that the method we proposed could generate realistic near infrared images and lay a solid foundation for the subsequent near infrared scene simulation of different ground objects on different weather conditions and at different time of the day.