Research On Point Cloud Registration Algorithm For Hip Arthroplasty

With the continuous development of modern computer vision and graphics,visual positioning and alignment techniques are widely used in the medical field.In modern minimally invasive artificial hip arthroplasty,the precise control of the force and angle of grinding of the acetabular fossa is one of the challenges to be solved intraoperatively.To solve this problem,many technicians have begun to investigate the use of joint replacement robots to perform surgery instead of the surgeon.In clinical surgery,the patient’s incision must not be exposed to air for a long time,therefore,the positioning accuracy of medical equipment,the effect of medical image pre-processing and the length of surgery are required to minimize the trauma suffered by the patient.In view of the fact that human skeletal joints are different from other soft tissues and their 3D structure information is more easily captured directly from the outside by sensors such as cameras,this paper adopts 3D point cloud alignment,a technique with small data volume,easy data acquisition and fast processing speed,to control the rapid and accurate positioning of the hip joint by the robot.The network similarity-based hip point cloud alignment is divided into three stages: point cloud acquisition stage,point cloud filtering(pre-processing)stage and point cloud alignment stage.I have done the following work in these three stages in the research process:(1)First,in the point cloud acquisition stage,when the doctor acquires the coordinates of the marker ball on the probe,the image of the probe captured by the camera is easily disturbed by light and electromagnetic waves from various devices in the surgical environment,resulting in blurring and distortion of the image.Since the traditional Hough transform circle detection algorithm cannot achieve the ideal detection effect when processing such images,this thesis proposes an algorithm for detecting circles based on edge grayscale and conducts detection experiments on several groups of images,and the results show that the average error between the measured circle coordinates and the true value of this algorithm is about 0.5 pixels,which is better than other traditional circle detection algorithms,indicating that this method can effectively detect the circle coordinates accurately from the probe images.(2)Secondly,in the point cloud filtering stage,the traditional clustering and distance-based filtering algorithm has some defects,for example,the process of point cloud filtering is easy to mistakenly divide the normal values,which leads to the change of the point cloud structure,and the filtering effect is not good for the high-frequency region of the model.To address these problems,this thesis proposes a point cloud noise reduction method based on optimized Principal Component Analysis(PCA)and guided filtering.The experimental results show that the proposed algorithm can successfully remove the redundant noise,and the ratio of filtered model to noise-free model points can reach 96.99%,while it can maintain the point cloud features better,which is better than several other traditional methods in terms of filtering effect.(3)Finally,in the point cloud alignment stage,the traditional iterative closest point(ICP)point cloud alignment algorithm deviates from the true value when selecting the initial point pair,which easily leads to too many iterations and falls into the local optimal solution.To address the above problems,this thesis proposes a network node similarity-based screening method for the initial point pairs,which firstly organizes the point cloud to be aligned into a network structure,secondly selects the point pairs closest to the true value by the network node similarity measure for iterative optimization,and finally completes the alignment by optimizing the error function.In the simulation alignment experiments,the real point cloud data extracted by the probe is compared with the CT-generated point cloud data,and the ICP alignment algorithm is compared using ICP and its different variants.The experiments show that the improved algorithm can accurately filter the initial point pairs on the hip joint surface for alignment and achieve fast and accurate alignment in any position of the two sets of point clouds extracted by CT and binocular camera,respectively,during the surgery.The optimized alignment algorithm improved the error by about 7% on average compared with the traditional point cloud alignment algorithm,and the alignment speed was 2.31 s,which was better than the traditional point cloud alignment algorithm.