Algorithm based on super low resolution video image processing

在目前的视频图像处理领域，人们倾向于尽可能地利用计算资源，而不是试图降低给定运动估计算法的计算复杂度来实现实时图像处理。本研究以具有非常简单的神经系统、能够实时控制其飞行动作的昆虫为研究对象，从理论上分析了它们的计算模型，即所谓的初级运动检测器（EMD）及其衍生的精细Reichardt检测器（ERD），以开发一种低计算复杂度的运动估计算法，该算法可以实时输入非常低分辨率的视频图像。在我们的研究中发现：(1)估计结果是这些模型中包含的滤波器参数的函数；(2)通过优化滤波器参数可以大大提高估计精度；(3)这种优化可以通过解析和数值相结合的方法来完成。通过在软件程序和硬件资源中实现我们基于上述数值模式开发的算法的性能进行了检验。该算法可以成功地估计运动，虽然在某些有限的条件下，在实时利用非常低分辨率的视频图像序列。

In the present video image processing community, people tend to utilize computational resources as much as possible rather than try to reduce the computational complexity of a given motion estimation algorithm for realizing real-time image processing. In our research, insects with very simple neural systems, capable of controlling their flight maneuvers in real time are focused and their computational models, so called an elementary motion detector (EMD) and its derivative an elaborated Reichardt detector (ERD), are theoretically analyzed for developing an motion estimation algorithm of low computational complexity to which very low resolution video images are fed in real time.In our research, it has been found that (i) estimation results are a function of filter parameters included in these models, (ii) estimation accuracy can be improved greatly by optimizing the filter parameters, and (iii) such an optimization can be done in a combination of analytical and numerical procedures. The performance of our developed algorithms based on the above numerical modes was examined by implementing them in software programs and hardware resources. The algorithms can estimate motion successfully, although in certain limited conditions, in real time utilizing very low resolution video image sequences.