应用NAR运动估计的序列帧间匹配技术

摘要

For the new method of high resolution remote sensing combined with push-frame and digital TDI post-processing, the step of inter-frame registration in the image post-processing process spends a lot of computing resources and time. This paper firstly introduces the application pattern of push frame imaging and digital TDI. By analyzing the spatial correlation characteristics and image registration mechanism of the original image sequence in the push-frame imaging mode, this fast inter-frame registration technique is proposed and experimentally verified. The results show that the fast inter-frame registration technique of NAR (Nonlinear Auto-Regressive models) neural network motion estimation can significantly improve the real-time performance of digital TDI post-processing, the efficiency of calculation by more than 70%, and the registration accuracy. This technology is highly generalized and versatile. Imaging data with different ground resolutions and different orbital heights can apply this technique to achieve a fast and accurate inter-frame registration.%对于面阵相机推帧成像与数字TDI后处理相结合的对地高分辨率遥感新方式,在图像后处理过程中的帧间匹配步骤占据了大量运算资源与时间.文章首先简要介绍了推帧成像与数字TDI的应用模式,通过分析推帧成像模式下原始图像序列的空域互相关特性与图像匹配机理,提出了此快速帧间匹配技术,并进行了仿真试验.结果表明应用非线性自回归神经网络运动估计的快速帧间匹配技术可显著提高数字TDI后处理的实时性,使计算效率提升达70%以上,同时提高匹配精度,减少误匹配.该技术泛化力强,通用度高,不同地面分辨率、不同轨道高度的空间相机数据均可应用该技术进行快速且精确的机内或地面帧间匹配.