Research On Doctor's Head Positioning Method By Kinect

With the development of augmented reality technology in the medical field,the surface projection method has been applied in minimally invasive surgery.One of the key technologies of this method is to accurately position the doctor's head to perform the image in the doctor's vision real-time display.At present,there are many head positioning technologies,but these technologies are poor in terms of accuracy,realtime performance,and robustness.There are many defects in practical applications.Therefore,this paper focuses on face detection,feature point extraction,and head positioning.Three aspects of modeling are used to optimize and improve the head positioning technology.Firstly,a face detection algorithm based on RGB-D images and Ada Boost is designed for the face wearing a mask.Use a Kinect camera to obtain RGB images and Depth images,use Depth data to segment possible face areas in the RGB images and binarize the images,perform median filtering and morphological opening and closing operations on the binarized images.The size of the rectangular frame is used to filter the candidate face area.Using this candidate area for Ada Boost face detection,a higher detection rate is obtained than the traditional Ada Boost algorithm.Secondly,the real-time location of feature points using the CLM algorithm may result in inaccurate positioning and disappearance of feature points.A Kalman filterbased real-time location algorithm for feature points is proposed.Use the CLM algorithm to locate the feature points of the face in the rectangular frame area marked by face detection,and introduce the Kalman filter to track the trajectory of the feature points.The problem of matching the measured and observed values during tracking is eliminated With the matching method,the obtained feature points have better continuity and higher accuracy than when using the CLM algorithm alone.Finally,in order to verify the accuracy of the head localization results,an experimental method was designed using the Bi Wi Kinect dataset as a test image.The three-dimensional feature points of the left and right pupils and the nasal point are selected to establish a geometric model to calculate the head rotation information Yaw,Pitch,and Roll.The normal of the plane where the three points are located is used as the perspective vector to calculate the translation information X,Y,Z.The face without a mask is verified using the Bi Wi Kinect dataset,and the face of a mask is verified using design experiments.It is concluded that the algorithm in this paper has higher accuracy and better real-time performance for head positioning with face mask.