基于SIFT及射影变换的多摄像机目标交接

Object handoff in multi-cameras based on SIFT and homograph

运动目标正确交接是多摄像机视频监控中的关键,视野分界线是解决目标交接的有效工具。不需标定摄像机参数,提出了一种利用尺度不变特征变换(SIFT:scale-invariant freatures transform)及射影变换实现目标交接的算法。首先使用SIFT算法在不同视角拍摄的图像间自动生成匹配的特征点,由空间共面的特征点及其相应匹配点生成图像间的单应变换矩阵。然后由图像边界点及单应矩阵计算摄像机视野(FOV:field of view)分界线。最后利用目标位置信息及射影变换实现目标正确交接。实验结果表明本文的方法具有有效性和鲁棒性。

To establish the correspondence between moving objects is a key problem in multi-camera surveillance,and field of view（FOV） line is an efficient tool to resolve the consistency in labeling objects.In this paper,we propose an algorithm that realizes object handoff by using scale-invariant features transform（SIFT） and homograph,without knowing the camera calibration information.Firstly,by using SIFT algorithm,matching points are automatically generated between two images sharing a joint region.We chose the matching points which are coplanar in space.These points are then used to compute the homography matrix of the two images.Then,the camera FOV lines are obtained by using the homography matrix and boundary points of the images.Finally,we realize the object handoff using the position of object and homograph.Experimental results show the accuracy and robustness of our method.