Method and device for extracting light bar center based on multi-scale

Abstract

The invention discloses a method for extracting the center of a light bar based on multi-scale, which includes performing noise smoothing processing on an image to obtain a smooth image; processing the smooth image through a skeletal method to obtain the initial center point of the light bar and the corresponding initial method line direction; perform grayscale Gaussian function fitting on the cross-section of the light strip along the initial normal direction of the light strip to obtain the width of the light strip at each position of the light strip; determine the first convolution at each position according to the width of the light strip Gaussian kernel mean square error σ0; according to the first convolutional Gaussian kernel corresponding to σ0, use the Hessian matrix method to obtain the pixel-level candidate points in the center of the light bar; use each candidate point as a base point to obtain the sub-pixel coordinates of the light bar image features; connect Each light bar center point forms the light bar center. Using the method for extracting the coordinates of the center point of the laser light bar of the present invention, by selecting the optimal Gaussian kernel mean square error at each position of the light bar, the coordinates of the center point of the light bar are extracted, which has the characteristics of high precision, good versatility, and strong anti-interference ability .

Description

[0001] technology neighborhood [0002] The present invention relates to the field of image processing, in particular to a method and device for extracting the center of light bars based on multi-scale. Background technique [0003] In the online structured light measurement system, the accurate extraction of the center of the light strip is one of the key factors affecting the accuracy of the entire measurement system. Common methods for extracting the centerline of light bars include: [0004] Edge method and threshold method: the algorithm is simple, the operation speed is fast, but the accuracy is low. [0005] Extremum method: Gaussian or parabolic fitting is performed on the cross section of the light strip, and the sub-pixel position of the center of the light strip is obtained by calculating the extreme point. This method is easy to extract the center of the light strip due to uneven gray distribution of the light strip and noise interference. The accuracy is not hig...

AI Technical Summary

Problems solved by technology

[0005] Extremum method: Gaussian or parabolic fitting is performed on the cross section of the light strip, and the sub-pixel position of the center of the light strip is obtained by calculating the extreme point. This method is easy to extract the center of the light strip due to uneven gray distribution of the light strip and noise interference. The accuracy is not high;

[0006] Center of gravity method: The center of gravity method can reduce the error caused by the asymmetry of the gray distribution of the light strip, but due to factors such as the surface and structure of the measured object, the curvature of the light stripe image changes greatly. In order to improve the accuracy of the center of the light strip, First, the Hessian matrix, Sobel gradient operator, and direction template need to be used to determine the normal direction of the light strip, resulting in a complex algorithm and a large amount of calculation;

[0007] Steger method: Dr. Steger C of Germany uses the Hessian matrix to obtain the normal direction of the light strip in the image, and then finds the extreme point in the normal direction to obtain the sub-pixel position of the light strip centerline, which has high precision, good robustness and application Wide range and other advantages; however, for laser light bar images with high reflection and sharp changes in light bar thickness, it is still impossible to achieve high-precision extraction of the light bar center; and a large number of convolution operations in the operation process lead to slow calculation speed

Images