基于Adaboost首帧检测的时空上下文人脸跟踪算法

Spatio-Temporal Context Face Tracking Algorithm Based on Adaboost First Frame Detection

针对时空上下文算法首帧需进行手动框选及选择偏差带来后续跟踪干扰的不足,提出利用Adaboost算法进行首帧检测,引入Kalman预测机制辅助时空上下文算法进行跟踪。当出现遮挡、抖动等问题时,保证跟踪稳定地进行,提高算法鲁棒性。在Shelter1等3组公共数据集上进行对比实验的结果表明,该算法能实现首帧自动检测功能,后续跟踪算法的鲁棒性及跟踪效果也得到明显提升。

Aiming at the problem that the first frame needed to be manually selected and the subsequent tracking interference was caused by selection deviations in the spatio-temporal context tracking algorithm,we proposed to use Adaboost algorithm to detect the first frame and introduce Kalman prediction mechanism to assist spatio-temporal context tracking algorithm.When occlusion,jitter and other problems occurred,it could ensure the stability of tracking and improve the robustness of the algorithm.Comparative experiments were carried out on three public data sets,such as Shelter1.The experimental results show that the proposed algorithm can realize the automatic detection function of the first frame,and the robustness and tracking effect of the subsequent tracking algorithm are also significantly improved.