Abstract: In strapdown imaging homing guidance system, the change of the distance between missile and target and the change of imaging visual angle may cause the rotation and scaling changes of image. The classical mean shift tracking algorithms can not robustly track a target which is under rotation and scaling. A scaling and rotation changing adaptive mean shift target tracking algorithm is proposed. The selected elliptic tracking region model and candidate tracking region model are weighted. Meanwhile, the moment characteristic and Bhattacharyya coefficient of weighted image are used to precisely estimate the actual area of target. Then a covariance matrix which can express the characteristic of image in tracking window is constructed using the estimated area and the second order center moment of weighted image. The relationship between the ellipse area and the eigenvalues of covariance matrix is established by singular value decomposition, the length and orientation of principal axis of ellipse are figured out, and the adaptive change of tracking window is realized. Test results indicate that this algorithm retains the high accuracy and real time of the classical mean shift algorithm, and at the same time, extends the adaptive ability of mean shift algorithm for the changes in scale and rotation of target.

Key words: ordnance science and technology, image guidance, strap-down, mean shift algorithms, moment feature, adaptation