Rapid detection method of sub-pixel precision industrial object

Abstract

The invention relates to a rapid detection method of a sub-pixel precision industrial object, comprising the following steps of: firstly, smoothening an image to be detected or a template image and extracting edges; taking coordinate information as an edge point; then, calculating a normal vector and a tangent vector of points of inner and outer outlines of objects on two images and converting the normal vector and the tangent vector into angle information; extracting Shape Context information from the edge points of the inner and outer outlines of the objects of the images; utilizing the angle information to correct a Shape Context descriptor of each point to obtain an improved descriptor; adding information including positions and the like into the improved descriptor to obtain characteristic vectors of the edge points; and finally, utilizing an ICP (Inductively Coupled Plasma) iterative algorithm to repeatedly match respective characteristic point and calculating an inverse transformation matrix to obtain sub-pixel position information of each point on the whole inverse transformation matrix and the outlines of the objects to be detected. The rapid detection method of the sub-pixel precision industrial object has the advantages of easiness of being realized, strong robustness, high precision, strong instantaneity, unchanged rotation and the like.

Description

technical field [0001] The invention relates to the technical fields of image processing and computer vision, in particular to a rapid detection method for sub-pixel precision industrial objects. Background technique [0002] Sub-pixel precision positioning information such as the position and angle of an object is an important link in an object detection system. Usually, in order to ensure that the object to be detected in a real scene obtains high-precision position and angle information, the positioning accuracy of the object in the image must reach the sub-pixel level. Since the position information provided by a single pixel in the image is only at the pixel level, obtaining sub-pixel position information needs to be realized by using other points in the neighborhood around the pixel. There are two ways to improve the positioning accuracy: [0003] 1. When extracting features from objects, more complex descriptors are used, which can be distinguished at the sub-pixel l...

AI Technical Summary

Problems solved by technology

In the case of sparse texture, the neighborhood features are often unstable, and the extraction and description of feature points are time-consuming, which often cannot meet the real-time requirements.

Another simple feature, such as corner points, edges and other information, will cause difficulty in subsequent matching, and the accuracy of the position information of feature points after scaling and multi-dimensional transformation will also be greatly affected.

[0019] To sum up, the current common high-precision positioning methods cannot adopt any method, and they cannot always be applied to rich and diverse industrial objects, nor can they be used in special cases (sparse textures, multiple symmetries, uneven illumination, noise, etc.) Large, etc.) still maintain a certain degree of high precision

Images