一种基于深度学习的视频单应性估计算法

A Deep Video Homography Estimation Algorithm

在视频场景下,针对图像帧序列,实现一种基于深度学习的单应性估计算法。传统单应性估计算法主要经过特征点提取和误匹配剔除等步骤,区别于传统算法,运用深度学习算法,使用类似于VGG架构的回归网络,对具有连续性的视频帧序列,进行单应性矩阵估计。

In a video scene,a homography estimation algorithm based on deep learning is implemented for image frame sequence. The traditional homography estimation algorithm mainly goes through the steps of feature point extraction and mismatching elimination. Different from the traditional algorithm,it applies the deep learning algorithm and using the regression network similar to VGG structure to estimate the homography matrix for the video frame sequence with continuity.