基于运动信息和相位一致性的运动目标提取

提出一种新的时空结合的运动目标分割方法。该算法首先在累积帧差图像上使用改进的高阶统计(HOS)算法检测出运动区域,用最大连通区域表示初始运动目标模板,利用基于Log-Gabor复小波变换的相位一致性边缘检测算法提取出每一帧的单边缘信息,结合初始运动模板与空域边缘图像得到更精确的运动目标模板,最后结合原图像分割出运动对象。实验结果表明该算法能从对比度较小和噪声较大的视频序列中精确地提取出运动目标,且不受光照变化的影响。

Moving Object Extraction Based on Motion Information and Phase Congruency

A new algorithm for extracting moving object was proposed based on spatio-temporal union information. First, motion areas were detected on accumulative frame-difference images based on an improved high-order statistics (HOS) algorithm. After selecting the largest connected area, a rough motion mask could be obtained. Second, a phase congruency edge detection algorithm based on Log-Gabor complex wavelet transform was used to extract the one-pixel width edges of every frame. Finally, a fine motion object mask was obtained by integrating the rough motion mask and spatial edge image. The moving object could be extracted through combining the fine motion mask with the original image. The experimental results show that the proposed algorithm can detect moving object accurately from the video sequence with low contrast and serious noise, and it is insensitive to illumination variation.