To realize the quality measurement of the spot facing of the aircraft skin, a structured light 3D vision inspection system is employed in this paper. Due to the high reflection of the workpiece and the influence of the noise, the system mainly solves the problem in the actual project. The acquisition of 3D point cloud data is partially missing. At the same time, due to the complexity of the surface parameter equation on the surface of the countersunk hole, the measurement of the normal deviation angle is difficult and accurate. Aiming at the above problems, the establishment of a spatial cone parameter model based on the orthogonal projection method for quality inspection of countersunk workpieces, and a mathematical model for optimizing the normal deviation angle parameters is proposed to improve the inspection accuracy. First, most points of the workpiece scanned by a point cloud camera are disordered and discrete. In the process of devising, normal estimation, and surface fitting of the original point cloud, it is necessary to operate based on the neighborhood. If there is no efficient auxiliary data structure, and if it is convenient to all points in space to have the k closest distances to a given data point index point, this time complexity will be very large for a large number of data points. KD tree, or k-dimensional tree, is a data structure employed in computer science. It mainly divides data through dimensions and efficiently manages high-dimensional data. In this paper, KD-tree is used to determine a spatial index structure index of scattered points.The first crucial link in the quality inspection process of 3D point cloud spot facing is to denoise the original point cloud. The fundamental purpose of this operation is to remove the influences of point cloud noise and provide high-precision and high-quality 3D data for subsequent workpiece quality inspection. In this experiment, when a surface structured light camera is used to obtain the three-dimensional point cloud data of the workpiece, due to the complexity of the actual working environment, there are various errors in the equipment obtained by the point cloud, which can generate a large number of noisy point clouds. These noisy point clouds are 3D data irrelevant to the detection target. And these noisy data are scattered and disordered spatial point clouds. Noise may also be generated due to external interference, line of sight occlusion, reflection and diffraction characteristics of some metal workpiece surfaces, obstacles, and other factors. To ensure that the quality parameters of the countersink can be accurately detected, this paper adopts a bilateral filtering noise reduction method to reduce the influence of outliers and noise. Secondly, the RANSAC algorithm is used to segment the upper surface, extract the upper and lower boundaries according to the characteristics of the centralized distribution of the boundary point cloud on one side, and the plane parameters are fitted. According to the orthogonal projection, the cone angle satisfies the Fourier transform function relationship on the plane XOY or the plane XOZ, the cone angle and cone point of the spot-facing hole are calculated, and an accurate spatial cone mathematical model is established. Finally, through data fusion between the upper and lower boundary planes of the spot facing and the spatial cone mathematical model, the parameters of the spot facing to be detected are obtained. When the spatial conical axis and Z axis have ceviation angle, the optimal mathematical model of normal deviation angle is proposed. For the surface parameter equation of surface countersink is too complicated, the problem of low normal precision fordetection countersink is avoided. Furthermore, to verify the algorithm performance, comparisons are made with machine vision and mobile least squares surface fitting methods. The experimental results show that the algorithm has higher accuracy and better anti-noise performance.