Light bar positioning method based on convolutional neural network

Abstract

The invention discloses a light bar positioning method based on a convolutional neural network, belongs to the field of visual measurement, and relates to a light bar positioning method based on a convolutional neural network. The method comprises the following steps: firstly, establishing a laser-assisted binocular vision measurement system, acquiring an image containing a laser light bar throughthe laser-assisted binocular vision measurement system, acquiring a large number of light bar images to make a data set, and classifying the data set; then, determining a convolutional neural networkstructure and parameters, and training the convolutional neural network structure and parameters by utilizing the data set; and finally, achieving light bar positioning based on non-feature filteringaccording to a training result. The method meets the requirements of measurement accuracy and real-time performance, and overcomes the influence of conditions of light bar bending, light and shade part overlapping and the like in a complex measurement environment. Accurate positioning of the laser light bar is achieved, and the method has the advantages of being accurate in positioning, high in robustness and the like.

Description

technical field [0001] The invention belongs to the field of visual measurement, and relates to a light bar positioning method based on a convolutional neural network. Background technique [0002] With the advancement of industrial production automation, visual measurement systems have been widely used in industrial processing measurement, product quality inspection and other fields. As the main measurement characteristic information in industrial measurement, laser light strip has been applied in industrial visual inspection fields such as welding seam inspection and surface scanning of aeronautical structural parts because of its high brightness, excellent linearity and good stability. The positioning of the laser light bar in the image determines the feature processing area and directly affects the robustness and accuracy of the image processing process. The positioning methods used in industrial measurement are mainly template matching method and feature-based template...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problems of poor robustness and low efficiency in the traditional positioning method for the line laser light strip image with uneven light and shade and complex image background involved in the appeal method, the present invention invented a convolutional neural network (CNN)-based light bar positioning method

This method meets the requirements of measurement accuracy and real-time performance, and overcomes the influence of twists and turns of light bars and overlapping of bright and dark parts in complex measurement environments.

Images