融合时域特征识别的前景检测算法

Foreground Detection Based on Feature Recognition in Time-domain

前景检测是计算机视觉领域的重要研究内容之一,动态背景对实际检测效果有很大的影响.针对复杂场景下的前景目标检测进行了深入的研究,提出一种融合时域特征识别的前景检测算法.首先对场景建立稳定的背景模型,通过该模型初步识别出可能存在的前景区域,然后利用小波变换提取此区域的时域变化特征,通过对时域特征分析进一步区分动态背景与真实运动目标,从而降低动态背景对检测效果的影响,最后利用形态学和连通性分析对结果进行后处理.实验结果表明,本文算法可以有效的降低动态背景区域的误检测率,提高整体检测效果.

Foreground detection is a critical and fundamental part in computer vision. Dynamic background has a great effect on the detection results. This paper presents a new approach for foreground detection based on feature recognition in time-domain. First,establish a stable model and find out the possible foreground area,then extract the feature of temporal variation in the area by w avelet transform,then identify the dynamic background by feature classification. Finally,morphology and connectivity analysis w ill be used on the detection results. Experimental results demonstrate that the proposed approach leads to a low er false positive rate and a better result for foreground detection.