基于计算摄影的真彩高清夜视成像方法研究

Night vision imaging technology is widely used in military reconnaissance, night combat, security monitoring and many other fields. Traditional low-light night vision and infrared night vision can obtain night scene texture and thermal radiation images, but it is difficult to obtain color images. How to achieve true-color and high-definition night vision images is a challenging task in the field of night vision imaging technology. Recently, scholars have conducted a lot of research on the fusion of infrared and visible images. However, how to improve color night vision imaging performance - resolution, signal to noise ratio, color fidelity, dynamic range and other core issues, is usually underestimated. Therefore, this project intends to apply the advanced methods and theories of computational photography to enhance the night vision imaging signal-to-noise ratio through the method of motion measurement, and obtain high-resolution, high-definition, true-color night vision images through image fusion methods, and improves night vision imaging dynamic range through the guidance of infrared images, thus providing technical support for high-performance infrared-visible fusion color night vision imaging.

夜视成像技术被广泛用于军事侦察、夜间作战、安防监控等诸多领域。传统微光夜视与红外夜视，可以获取夜间场景纹理与热辐射图像，但难以获取彩色图像。如何实现真彩高清夜视成像是夜视成像技术领域研究的难点。近期，学者们在红外与可见光融合夜视成像方面开展了大量研究中，但是，对如何提升彩色夜视成像性能——分辨率、信噪比、颜色保真度、动态范围等核心问题，却鲜有研究。为此，本项目拟借鉴计算摄影中先进方法和理论，在现有红外与可见光器件水平条件下，通过基于运动测量的时空域图像去噪增强夜视成像信噪比，通过基于混合成像的高分辨灰度图像与低分辨率彩色图像融合获得高分辨率真彩色夜视图像，通过红外图像辅助实现宽动态夜视成像，为真彩高清夜视成像应用提供技术支撑。

夜视成像技术被广泛用于军事侦察、夜间作战、安防监控等诸多领域。传统微光夜视与红外夜视，可以获取夜间场景纹理与热辐射图像，但难以获取彩色图像。如何实现真彩高清夜视成像是夜视成像技术领域研究的难点。近期，学者们在红外与可见光融合夜视成像方面开展了大量研究中，但是，对如何提升彩色夜视成像性能——分辨率、信噪比、颜色保真度、动态范围等核心问题，却鲜有研究。为此，本项目拟借鉴计算摄影中先进方法和理论，在现有红外与可见光器件水平条件下，通过搭建的多传感器数据采集平台，获取了包括红外图像、彩色图像、灰度图像、深度图像、以及惯导单元提供的姿态信息。基于采集的大量数据进行图像增强、颜色恢复和运动去模糊的多层神经网络训练，取得了高分辨率真彩色夜视成像的初步结果，为下一步在夜视图像设备的研发奠定了技术基础。