光流计算中的时域微分估计误差分析与逐次逼近计算模型

Error analysis of temporal differential estimation and successive approximation calculation model in optical flow estimation

光流计算是实现视频序列中运动补偿滤波和视频对象三维运动分析的基础,也是检测与跟踪可见光图像和红外图像中运动目标的重要前提。而目前各种光流计算方法大多是基于一阶时空梯度的技术,都没考虑时域微分估计误差对计算准确性的影响。分析了光流计算中产生时域微分估计误差的各种因素,提出了光流的逐次逼近计算模型。通过位移预补偿方式减小了时域微分估计中的不准确性,极大地提高了光流计算精度。理论分析和大量对比实验证明,光流的逐次逼近计算模型比通常的光流计算方法具有更高的准确性。

The optical flow estimation is the basis of motion compensation filter in the video and 3-dimensional motion analysis of video objects. It is also an important primise for detecting and tracking of the moving objects in the infrared image. In general, the existing optical flow estimation methods are almost based on first-order spatiotemporal gradient. However, the error of the optical flow estimation derived from the error of temporal differential estimation is not taken into account in above methods. The factors, which introduce the error of temporal differential estimation are, analyzed. The successive approximation calculation model for the optical flow estimation is put forward. The displacement precompensation is utilized to reduce this kind of error, which greatly improves the accuracy of the optical flow estimation. The theoretical analysis and many comparative experiments have demonstrated that the optical flow derived form the successive approximation calculation model is better than that from the common methods.