摘要
在通过特征点做运动估计时,经常会出现特征点集中于图像部分区域的问题。通过分框统计MORAVEC特征点来计算各分块内的特征点的方法,对视频图像进行帧与帧间匹配,再结合RANSAC误匹配剔除的方法能有效解决该问题。通过对计算原始SENSOR图像中中间区域部分分块并预留原始图像的边缘部分,统计正确的匹配运动矢量,对运动矢量值加总并用惯性滤波的方法进行平滑拟合进而得到每帧图像的运动补偿值。通过该方法得到的特征点匹配对分布的区域明显比不采用该方法分布的区域广,而且该方法在SOC上能快速运算。
When the feature points are used to estimate the motion,the problem that the feature points will be concentrated in the region of the image often occurs. The method of calculating the feature points in each block by dividing the MORAVEC feature points to match the adjacent two frames of the video images,and then use the method of RANSAC mismatch elimination can effectively solve the problem. Based on the calculation of the middle area of the original SENSOR image grid and edge part reserved for the original image matching,statistics the right motion vector,the value of motion vector method and the total inertia filtering smoothing fitting and motion compensation values of each image. The distribution area of feature points matched by this method is obviously wider than that without the method. Moreover,this method used in SOC can be calculate fastly.
出处
《电子科技》
2017年第11期48-52,共5页
Electronic Science and Technology
基金
国家自然科学基金(61401140)
