混合Adaboost算法与Kalman滤波的头势人机交互

摘要

在基于头势的智能轮椅的人机交互中,头势识别是人机交互的重要研究内容.提出一种新的用于控制智能轮椅的头势识别方法,通过Kalman滤波器预测由Adaboost算法检测到的唇部窗口在下一帧图像可能出现的位置,进行下一帧图像的唇部检测,进而通过唇部窗口的位置与整个检测窗口中一个固定的矩形窗口进行比较确定头势.Kalman滤波器的使用克服了单纯使用Adaboost检测算法检测每一帧待检测窗口时将整个窗口的所有的位置都要搜索的问题,提高了唇部检测时间及检测精度,使智能轮椅在被头势控制过程中运动连续、正确.%Head gesture is an important intelligent human-wheelchair interaction approach, and the head gesture recognition is an important aspect of the interactions. A new head guests recognition to control the intelligent wheelchair presents in this paper, Kalman filter forecast the lips position detected by Adaboost algorithm may be appeared in the next frames first, and then detect the lips in the next frame. Compare the lips window position with a fixed rectangle to confirm the head gesture correspondingly. Kalman filter overcome detect all the possible lips position by just use the Adaboost algorithm in every frame, greatly improve the lips detection precision and reduce the detection time, solve the wheelchair's time delay during the head guests controlling.