Piston pose estimation method based on improved CVFH and CRH features

Abstract

A piston pose estimation method based on improved CVFH and CRH features, includes off-line processing and on-line recognition. The off-line processing firstly calibrates Kinect, then automatically renders the piston CAD model to generate multi-view point cloud, calculates its point cloud features, and finally generates off-line template library by indexing the point cloud and features. On-line identification is to estimate the position and attitude of pistons stacked randomly, firstly, the piston point cloud is collected and corrected by using the calibrated Kinect, then the point cloud is preprocessed and segmented, for the segmented point cloud, the feature of the point cloud is calculated and initially registered with the template library, then the point cloud is precisely registered byICP algorithm with the initial value of the posture, and the precise posture of the point cloud is obtained. For the generated precise posture, the mismatch is eliminated by the hypothesis verification algorithm, and the piston posture estimation and output are finally completed. The invention has the advantages of low time complexity, convenient obtaining of matching template and high accuracy of pose estimation.

Description

technical field [0001] The invention relates to the technical field of object pose estimation of machine vision, in particular to a random stacking piston pose estimation method based on improved CVFH and CRH features. Background technique [0002] In the field of machinery manufacturing, the loading process has always been a relatively weak link. It consumes a lot of time, has low efficiency, and is highly dangerous. Most safety accidents occur in loading and unloading operations. Solving the problem of feeding automation is of great significance for reducing labor intensity of workers, improving production efficiency, realizing multi-machine tool management and ensuring safe production. At present, the loading process of the domestic piston production line is to manually place the piston according to the required position and posture, and then grab it with a robot arm, which cannot meet the requirements of an efficient and flexible processing line. The use of machine visi...

AI Technical Summary

Problems solved by technology

[0003] The structure and stacking form of the target object are simple, and no research and experiments have been conducted on real workpieces and completely random stacking. The conditions for method verification are relatively simple and cannot be realized in a real factory environment.

[0004] The acquisition of matching templates is complicated, and most studies cannot achieve the matching between the CAD digital model of the workpiece and the point cloud collected by the sensor. The matching templates are generally collected manually

This acquisition method is low in accuracy and time-consuming, and requires professional equipment such as a three-dimensional rotating platform and a laser scanner, as well as the cooperation of relevant technical personnel. After scanning, it is still necessary to calibrate and extract the attitude of each point cloud, which is a complicated process. Difficult to achieve rapid deployment of the system

[0005] The adaptability of the method is poor. In order to achieve fast and accurate detection and recognition, the design of many related algorithms is closely related to the unique characteristics of the captured workpiece. Once the workpiece is replaced, the effect of the algorithm will be greatly reduced, and the adjustment of parameters is very difficult.

[0006] The cost of the system is high, and the sensors and computing equipment used in related research and products are very expensive, such as high-precision 3D acquisition systems and industrial-grade servers. The cost of the entire system is very high, which will largely affect the enthusiasm of enterprises to introduce products

[0007] In summary, the current pose estimation method has the disadvantages of unrealistic experimental environment and research objects, complex matching template acquisition, poor method adaptability and high system cost.

Images