Woods Matthew, Katsaggelos Aggelos
Northwestern University, Department of Electrical Engineering and Computer Science (EECS), Evanston, Illinois 60208, USA.
J Opt Soc Am A Opt Image Sci Vis. 2013 Jan 1;30(1):102-11. doi: 10.1364/JOSAA.30.000102.
This paper presents a computationally efficient method for the measurement of a dense image correspondence vector field using supplementary data from an inertial navigation sensor (INS). The application is suited to airborne imaging systems, such as an unmanned air vehicle, where size, weight, and power restrictions limit the amount of onboard processing available. The limited processing will typically exclude the use of traditional, but computationally expensive, optical flow and block matching algorithms, such as Lucas-Kanade, Horn-Schunck, or the adaptive rood pattern search. Alternatively, the measurements obtained from an INS, on board the platform, lead to a closed-form solution to the correspondence field. Airborne platforms are well suited to this application because they already possess INSs and global positioning systems as part of their existing avionics package. We derive the closed-form solution for the image correspondence vector field based on the INS data. We then show, through both simulations and real flight data, that the closed-form inertial sensor solution outperforms traditional optical flow and block matching methods.
本文提出了一种计算效率高的方法，用于利用惯性导航传感器（INS）的补充数据来测量密集图像对应向量场。该应用适用于机载成像系统，如无人机，在这类系统中，尺寸、重量和功率限制了可用的机载处理量。有限的处理能力通常会排除使用传统但计算成本高的光流和块匹配算法，如卢卡斯-卡纳德算法、霍恩-申克算法或自适应十字模式搜索算法。相反，从平台上的INS获得的测量结果可得出对应场的闭式解。机载平台非常适合此应用，因为它们作为现有航空电子设备包的一部分，已经配备了INS和全球定位系统。我们基于INS数据推导了图像对应向量场的闭式解。然后，通过仿真和实际飞行数据表明，闭式惯性传感器解优于传统的光流和块匹配方法。
