基于SIFT特征匹配与K-均值聚类的运动目标检测

Moving object detection based on SIFT features matching and K-means clustering

运动摄像机情况下的运动目标检测是视频监控中的难点和热点问题。为了能够有效地检测出运动目标,根据视频中背景与运动目标的速度不同这一特点,提出了一个基于尺寸不变特征变换(SIFT)和K-均值聚类的运动目标检测方法。首先提取视频中相邻两帧图像的SIFT特征点并进行匹配,并计算匹配特征点的运动速度,最后将运动目标和背景上的SIFT特征点K-均值聚类分析,在单运动目标、多运动目标和带有摄像头旋转情况下做了实验。实验结果表明,提出的目标检测算法能够在运动背景下较好地检测到目标并保留稳定的目标局部特征,对于摄像机运动、摄像机旋转、亮度变化等影响因素具有较强的适应能力。

It is a difficult and hot topic in video surveillance to detect moving objects with moving camera. In order to detect moving objects effectively, according to the characteristics of the different speed between the background and moving target, a method was proposed based on Scale Invariant Feature Transform （SIFT） features matching and K-means clustering. The SIFT features of the two adjacent frames in the video were extracted and matched firstly. After that the velocity of the matched SIFT features were computed. Finally the K-means clustering method was used to analyze the SIFT features of the moving objects and background and experiments were done in the cases of single moving object and muhiple moving objects and when the camera was rotated. The experimental results demonstrate that, the proposed method can detect targets effectively and remain the stable local features of targets in moving background and have good adaptability to changing illumination and camera movement and rotation.