【摘要】 给出一种新的头部跟踪算法,算法综合运用可变形模板、光流及扩展Kalman滤波技术,有效融合了图像中目标的形状、运动及彩色特征,从而获得快速、鲁棒的头部跟踪结果。算法使用椭圆作为目标的形状模板以捕获目标边缘,基于彩色YCrCb信息以实现光流估计,由于采用模型约束的特征光流估计技术,故无需引入任何附加的平滑性约束,借助扩展Kalman(EKF)滤波器以整合目标的形状与运动信息。最后利用光流测量方程给出的误差测度及EKF给出的估计方差完成对虚假边缘点的判断与舍弃,进而保证对图像噪声、遮挡及伪边缘点具有良好的识别与处理能力。实例结果表明了该算法的有效性。更多还原

【Abstract】 A new head-tracking algorithm based on deformable models, optical flow and extended Kalman filter techniques is proposed, which combined the shapes, motion and color cues of objects in images in order to acquire a fast and robust scheme of head tracking. The algorithm used ellipse as a deformable templates to capture the edge of objects. With no needs of any special smoothness constraint, the optical flow is estimated by model-based optical flow method in YCrCb color space. By using of the extended Kalman filter, the measurements of shape and motion are naturally integrated to provide an effective fusion solution. An optical-flow based measurement error and the estimated covariance given by the EKF filter are used to detect and reject the contour sample points that correspond to noise, occlusions or spurious edge. Experiments results are presented to validate the algorithms.更多还原