PTZ视频监控的大场景形成方法研究

The formation approach of wide-area scene in PTZ video surveillance

针对PTZ摄像机跟踪监控系统,提出一种基于关键帧和S.peeded Up Robust Features(SURF)特征的图像拼接方法,以构建大的监控场景.利用局部梯度极值平均法对监控视频各帧的清晰度进行评价,剔除模糊帧,选择若干清晰帧作为关键帧进行拼接,以提高大场景的形成速度和准确度.对于图像拼接,采用效果近似于Scale Invariant Feature Transform(SIFT)方法,但计算量更小、速度更快的SURF方法来提取特征点、寻找匹配点对,并结合RANSAC和L-M算法求图像间的变换关系.实验表明,相对于SIFT特征,SURF特征大大提高了图像拼接速度和精度,且具有较高的鲁棒性.该方法在满足实时性的要求下,很好地构建了大场景图像.

In order to build a wide-area scene for video surveillance, a background mosaicing method based on key frames and SURF features was proposed. Fuzzy frames were eliminated through evaluating image sharpness with local maximum gradient average value. In this way, the accumulated error genera- ted during sequence rnasaicing was decreased. Simultaneity, the speed of wide-area scene formation was enhanced by reducing the number of frames. For image mosaicing, the algorithm of SURF with less op- erations and higher speed than SIFT was adopted to extract features and find matched features, then the mapping relationship between images could be acquired accurately using RANSAC and L-M algorithms. Experimental results show that the precision and speed of SURF were better than SIFT, as well as robust. This method could meet the needs of real-time property and build the wide-area scene scanned by the PTZ camera excellently.