为解决因边界效应导致相关滤波跟踪算法不够稳健及其不能适应尺度变化的问题, 提出了一种基于双模型的相关滤波跟踪算法。将目标跟踪分为位置预测和尺度预测两部分, 在位置滤波器模型进行位置预测阶段, 先通过对待测样本进行样本增强处理, 使得到的样本更符合实际场景。再通过交替方向乘子法进行位置滤波器的迭代求解, 最后得到估计的目标位置。在尺度滤波器模型进行尺度预测阶段, 通过在估计的目标位置处构建多尺度金字塔来训练尺度滤波器, 再求解得到目标的尺度, 将双模型得到的结果作为最终的跟踪结果。最后通过引入一个遮挡判据来判断是否更新模型以提高算法的鲁棒性。实验表明, 改进算法和经典的相关滤波跟踪算法相比, 在跟踪成功率上提高了18%, 在跟踪精度上提高了11%。在目标被遮挡、自身尺度变化时, 改进算法仍能稳定跟踪。

For the correlation filtering tracking algorithm is not robust enough and cannot adapt to scale changes due to the boundary effect, an improved correlation filtering tracking algorithm based on double model was proposed. The target tracking consisted of position prediction and scale prediction. In the position prediction stage, the samples were enhanced to make them more consistent with the actual scene. Then, the solution was obtained using the alternating direction method of multipliers, and the estimated target position was achieved. For scale prediction, a multiscale pyramid was constructed to train the scale filter, and then the target scale was acquired. The final tracking result was determined by both the target position and scale. Finally, an occlusion criterion was introduced to determine whether the model is updated or not. Compared with the classical correlation filtering tracking algorithm, the proposed algorithm boosts the tracking success rate by 18% and tracking accuracy by 11%. The algorithm can track the target stably even when the target is occluded and its scale changes.