【摘要】 提出了一种先进行空间定位再确定目标尺度的两级跟踪算法,有效地解决了meanshift算法对尺度变化的适应问题。该算法首先在当前帧对应位置进行降分辨率处理,并以基于增量试探的meanshift跟踪算法收敛点作为当前帧目标中心位置,进而利用对数极坐标变换的旋转、尺度不变性,对目标和候选目标分别进行对数极坐标映射,并通过求取最大归一化相关函数确定目标的尺度变化。跟踪实验表明,该算法可以有效的提高meanshift跟踪算法空间和尺度定位准确性。更多还原

【Abstract】 A novel two-stage method is presented for the efficient determination of target center and scale,improving the adaptability for scale changes.Firstly,the proposed scheme downsamples the correesponding parts of current frame,and employs the mean shift algorithm with successive bandwidth trials to derive the center of target candidate.Secondly,by using the scaling and rotation invariance of log-polar transform,the changes of object scale and rotation are converted into transformation in orthogonal coordinate;then the maximum 1-D normalized correlation coefficient is calculated to estimate the scale variance of object.The experiments on various videos show that the proposed method improves the performance of mean shift tracking algorithm with localization accuracy and adaptive bandwidth.更多还原