A key point detection method of industrial parts based on deep learning

Abstract

The invention discloses a method for detecting key points of industrial parts based on deep learning in the field of industrial vision, which specifically includes the following steps: S1: training a deep neural network for key point detection: constructing a deep neural network for key point detection consisting of three sub-networks, Each sub-network contains several 3*3 convolution kernels with a step size of 1 to 2, and multi-size feature map fusion technology is used between each sub-network for feature fusion; S2: output the image to be detected to the training Key point detection deep neural network, using the key point detection deep neural network obtained from training to detect key points, and then using the key points to match and calibrate industrial parts, the convolutional neural network structure applied to key point detection of bearing workpieces, the network structure Contains a novel feature map fusion technique and target loss function, which can effectively improve the accuracy of key point detection.

Description

technical field [0001] The invention relates to the technical field of industrial vision, in particular to a method for detecting key points of industrial parts based on deep learning. Background technique [0002] For the current key point matching technology based on traditional images, for image processing technology, key point matching technology is often needed to find the affine transformation relationship between two pictures. First, the corner detection algorithm is usually used to detect key points through the pixel gray value; then, each key point is described by using local descriptors, such as SIFT, ORB, etc. In this way, a one-to-one pairing relationship between key points and descriptors is formed. When looking for an affine relationship, first detect the key points and their descriptors on the current picture, and then use the descriptors to find the closest point of the key points in the two pictures to form a point pair. Through a large number of point pair...

AI Technical Summary

Problems solved by technology

[0003] However, there are defects: the traditional method is greatly restricted by the image quality, and different shadows, deformations, and rotations will have a great impact on the detection and descriptor of key points, which means that the traditional method is easy to form a wrong match, and the image transformation and Environmental changes are very sensitive, and the stability and robustness are not enough

Images