Research On Three-dimensional Detection And Recognition Methods For Complex Geometric Shapes

Geometric shape detection and recognition in the 3-D scene is always a hot topic in the field of computer image and pattern recognition.It has a stronger description ability and richer features for the complex scene.However,compared with contact or two-dimensional image detection methods,there are still some problems in practical application,such as data redundancy,a large amount of calculation,density change and so on.In this paper,a multi-view stereo parts detection method based on the light cut contour is designed for the surface of stereo parts which is difficult to deal with by common detection methods.First of all,according to different 3-D reconstruction methods,after investigating the current mainstream methods and considering the actual application conditions,the appropriate line laser sensor and multi-degree of freedom motion control platform are selected for 3-D surface reconstruction of 3-D parts.Also,to improve the overall accuracy of the system,this paper proposes an automatic calibration method based on the 3-D point cloud data,which compensates the error through the point cloud processing algorithm,effectively improving the reconstruction quality of the system provides a guarantee for subsequent identification and detection.Secondly,for the recognition and detection of complex surface parts,this paper compares and compares the existing algorithms,introduces the feature of SHOT(signature of histograms of orientations)as the feature description of point cloud,extracts and matches the features of target point cloud and measurement point cloud of parts respectively based on point cloud simplified sampling,and then performs secondary screening through hypothesis generation and hypothesis verification Finally,the target point cloud with the least matching error is selected.The CAD(Computer Aided Design)model information of the component is automatically matched according to the recognition of the measurement point cloud.In the aspect of defect detection of parts surface,the method of point cloud slice comparison is used to replace the 3-D comparison of a point cloud,and the interpolation method of point cloud contour is improved to calculate the overall error distribution of parts surface.At the same time,through density clustering and segmentation to further locate the defect location,improve the accuracy and speed of surface error detection of stereo parts.Finally,to verify the effectiveness of the algorithm proposed in this paper,a multiview 3-D scanning system for 3-D parts recognition and detection is built,and the measurement accuracy of the system is evaluated.The experimental results show that the plane distance error of the system can reach 5 μm,the sphere fitting error can reach 0.012 mm,and the average measurement accuracy of the system is 0.08 mm,which can meet the tolerance standard of common mold processing.Besides,this paper selects a kind of common 3-D electrode mold as a specific case.The experimental results show that the 3-D detection and recognition method proposed in this paper can identify different parts and calculate the degree and position of defects,which has certain robustness and achieves the goal of 3-D detection and recognition.