FMCW laser radar point cloud ghosting removal method based on edge detection algorithm

Abstract

The invention discloses an FMCW laser radar point cloud ghosting removal method based on an edge detection algorithm, and belongs to the technical field of image processing. The method comprises the following steps: firstly, acquiring difference frequency data generated by a to-be-measured object based on an FMCW laser radar system, and carrying out object distance calculation; constructing an object point cloud model by using each scanning angle and the corresponding distance information of the to-be-measured object to realize laser radar scanning; performing edge detection on an XY front view plane for constructing the point cloud model, calculating a gradient magnitude and a gradient direction of an image through a Sobel operator, and extracting and identifying a contour range of each object; and then according to the identified contour range, the gray value of each Z-axis plane is detected, and the layer with the highest gray value is selected as the real position of the object. The method is simple in implementation mode, convenient to operate, high in practicability and suitable for application and popularization.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a method for removing ghosting from FMCW laser radar point clouds based on an edge detection algorithm. Background Art

[0002] FMCW Lidar (Frequency Modulated Continuous Wave Lidar) is a radar system based on optical coherent detection technology, mainly used for high-precision distance measurement and speed perception. Compared with traditional pulse Lidar (Time-of-Flight Lidar), FMCW Lidar can achieve higher measurement sensitivity and resolution by processing frequency-modulated continuous wave signals, and has inherent anti-interference capabilities.

[0003] Currently, when FMCW lidar is performing high-speed scanning, the movement changes of the rotating mirror may bring additional measurement uncertainties, resulting in ghosting in the generated point cloud modeling map, making it difficult to accurately locate the object. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a FMCW laser radar point cloud ghosting removal method based on an edge detection algorithm, so as to eliminate the unstable distance layer of the modeling point cloud and realize the precise positioning of the object. At the same time, the relevant theory of applying the edge detection algorithm to the FMCW laser radar is proposed.

[0005] The technical problem proposed by the present invention is solved in this way:

[0006] A method for removing ghosting from FMCW laser radar point cloud based on edge detection algorithm is implemented based on FMCW laser radar system; the FMCW laser radar system includes a tunable laser, an optical beam splitter, an optical galvanometer, a collimating lens, an optical coupler, a balanced photodetector, a data acquisition card and a PC;

[0007] The laser emitted by the tunable laser is divided into two beams by an optical beam splitter, namely, the measuring light and the reference light; the measuring light is deflected by the optical galvanometer, and then collimated by the collimating lens before reaching the object to be measured; the object to be measured generates reflected light, which is coupled with the reference light in the optical coupler after passing through the collimating lens to generate a difference frequency signal; the balanced photodetector performs photoelectric conversion and intensity collection on the difference frequency signal, and the collected difference frequency data is transmitted to the PC through the data acquisition card, and the PC performs frequency resolution based on the collected difference frequency data and its corresponding galvanometer angle to obtain the distance information of the object to be measured;

[0008] The specific process is:

[0009] Step 1: The tunable laser performs laser beat frequency by continuously sweeping the frequency-sweeping light source, and the scanning speed of the optical galvanometer matches the frequency-sweeping speed of the light source to obtain a difference frequency signal corresponding to each scanning angle; frequency resolution is performed based on the difference frequency signal to obtain the distance information of the object to be measured; and an object point cloud model is constructed using each scanning angle and the corresponding distance information of the object to be measured to realize lidar scanning;

[0010] Step 2, extracting the front view of the object point cloud model on a plane perpendicular to the laser emission direction;

[0011] Step 3: Preprocess the extracted front view, including grayscale image conversion and filtering and smoothing; perform edge detection on the preprocessed image based on the Sobel operator to obtain the object contour;

[0012] Step 4: Use the object contour to compare the grayscale value of each image slice of the object to be measured in the laser emission direction, and select the object distance corresponding to the image slice with the highest grayscale value as the real position information of the object to be measured, so as to achieve ghost removal.

[0013] Furthermore, the wavelength of the tunable laser is λ=1550nm, the modulation bandwidth is B=1.09GHz, and the modulation frequency is F m =125KHz, the maximum detection angle is 15 degrees.

[0014] Furthermore, in step 3, the grayscale image conversion is expressed as:

[0015] Y=0.299×R+0.587×G+0.114×B

[0016] Among them, R, G and B are the pixel values ​​of the red, green and blue channels respectively, and Y is the grayscale value.

[0017] Furthermore, in step 3, a Gaussian filter is used to smooth the grayscale converted image.

[0018] Furthermore, in step 3, the horizontal gradient operator used to detect vertical edges in the Sobel operator is and the vertical gradient operator for detecting horizontal edges They are:

[0019]

[0020] Furthermore, in step 3, the Sobel operator is used to calculate the horizontal gradient G of each pixel. x and the vertical gradient G y , then the gradient magnitude G and gradient direction θ of the pixel are:

[0021]

[0022] Furthermore, in step 3, the edge position is accurately determined by non-maximum suppression; if the gradient amplitude of the current pixel is greater than or equal to the gradient amplitudes of the two neighboring pixels along the gradient direction, the pixel is retained as an edge point; otherwise, the pixel is suppressed; edge thresholding is performed, strong edges exceeding the set threshold are retained, weak edges connected to strong edges are retained, and the remaining weak edges are suppressed.

[0023] The beneficial effects of the present invention are:

[0024] (1) The method of the present invention innovatively uses an edge detection algorithm to process FMCW point cloud information, thereby determining the true position of the object.

[0025] (2) The method described in the present invention uses the Sobel operator to implement edge detection of the scanning point graph, which has higher accuracy and faster detection speed.

[0026] (3) In the method described in the present invention, the Sobel operator is used for the edge detection algorithm, and the relevant formula for edge contour extraction in the FMCW laser point cloud image is given, thus establishing a theoretical basis.

[0027] (4) The method of the present invention is not only simple to implement and easy to operate, but also highly practical and suitable for popularization and use. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the process of the method of the present invention;

[0029] Figure 2 The basic structure diagram of the FMCW laser radar used in the method of the present invention;

[0030] Figure 3 (a) is a front view of a plane perpendicular to the laser emission direction in the method described in the embodiment, Figure 3 (b) is the result image before and after edge detection;

[0031] Figure 4 The figures are spatial modeling effect diagrams before and after the ghosting is eliminated by the method described in the embodiment, wherein (a) is before elimination and (b) is after elimination. DETAILED DESCRIPTION

[0032] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0033] This embodiment provides a method for removing ghosting from FMCW laser radar point cloud based on an edge detection algorithm, and its flow chart is as follows: Figure 1As shown, the problem of ghosting in the constructed point cloud model and difficulty in accurately locating and modeling objects under high-speed rotating mirror scanning is solved. Let the laser emission direction be the Z axis, and the XY plane is perpendicular to the Z axis. The method described in this embodiment is based on the Sobel operator to perform edge detection on the XY front view plane for constructing the point cloud model, and then detect the grayscale value of each plane at different positions on the Z axis according to the identified contour range, and select the layer with the highest grayscale value as the true position of the object. In this way, accurate positioning of the object is achieved. The FMCW laser radar point cloud ghosting removal method based on the edge detection algorithm described in this embodiment is not only simple to implement and easy to operate, but also has strong practicality and is suitable for popularization and use.

[0034] The method described in this embodiment is implemented based on the FMCW laser radar system. Figure 2 As shown in the figure, the FMCW laser radar system includes a tunable laser, an optical beam splitter, an optical galvanometer, a collimating lens, an optical coupler, a balanced photodetector, a data acquisition card and a PC. The laser emitted by the tunable laser is divided into two beams of light through the optical beam splitter, namely the measurement light and the reference light; the measurement light is deflected by the optical galvanometer, and then collimated by the collimating lens before reaching the object to be measured; the object to be measured generates reflected light, which is coupled with the reference light in the optical coupler after passing through the collimating lens to generate a difference frequency signal; the balanced photodetector performs photoelectric conversion and intensity acquisition on the difference frequency signal, and the collected difference frequency data is transmitted to the PC through the data acquisition card, and the PC performs frequency resolution based on the collected difference frequency data and its corresponding galvanometer angle to obtain the distance information of the object to be measured.

[0035] The wavelength of the tunable laser in the structure adopted in this embodiment is λ=1550nm, the modulation bandwidth is B=1.09GHz, and the modulation frequency is F m =125KHz, the maximum detection angle is 15 degrees, and the distance between the detection target portrait and the optical lens is 6m.

[0036] The method described in this embodiment specifically includes the following steps:

[0037] Step 1: Perform laser beat frequency by continuous frequency sweeping of the frequency sweeping light source. The scanning speed of the galvanometer matches the scanning speed of the light source to obtain the difference frequency signal corresponding to each scanning angle. Perform frequency resolution based on the difference frequency signal to obtain the distance information of the object to be measured. Use the distance information corresponding to each scanning angle to construct an object point cloud model to realize lidar scanning.

[0038] Step 2: Extract the front view of the object point cloud model on the plane perpendicular to the laser emission direction, such as Figure 3 (a) shown.

[0039] Step 3: Perform image preprocessing on the extracted front view.

[0040] The first is grayscale image conversion: If the image is a color image, it is converted to a grayscale image. A grayscale image is the brightness value of each pixel and does not contain color information. The conversion to a grayscale image can be done using the following formula:

[0041] Y=0.299×R+0.587×G+0.114×B

[0042] Among them, R, G and B are the pixel values ​​of the red, green and blue channels respectively, and Y is the grayscale value. Image denoising will be performed afterwards: In practical applications, images are usually affected by noise, and Gaussian filters are usually used to smooth the images to reduce the impact of noise on gradient calculation.

[0043] The Sobel operator is used to perform edge detection on the preprocessed image.

[0044] Furthermore, the Sobel operator uses two convolution kernels to calculate the horizontal and vertical gradients of the image:

[0045] Horizontal gradient operator (used to detect vertical edges):

[0046]

[0047] Vertical gradient operator (used to detect horizontal edges):

[0048]

[0049] By convolving the preprocessed image with the two convolution kernels, the horizontal and vertical gradients of each pixel are calculated. The convolution operation can be performed using a standard image processing tool library (such as OpenCV).

[0050] Convolution operation: The convolution kernel performs a weighted summation with each local area of ​​the image pixel by pixel, and outputs the gradient value of each pixel in the image. For horizontal gradients, the Sobel convolution kernel identifies horizontal edges (such as horizontal lines with light and dark changes in the image). For vertical gradients, the Sobel convolution kernel identifies vertical edges (such as vertical lines with light and dark changes in the image).

[0051] Furthermore, through the horizontal gradient G x and the vertical gradient G y , the gradient magnitude and direction of each pixel can be calculated:

[0052] Gradient magnitude (indicating edge strength):

[0053]

[0054] This formula calculates the composite value of the gradient strength of each pixel in the horizontal and vertical directions. The larger the gradient magnitude, the stronger the edge of the pixel position.

[0055] Gradient direction (indicates the direction of the edge):

[0056]

[0057] This formula calculates the gradient direction of each pixel and is usually used for subsequent analysis of edge orientation in the image.

[0058] Furthermore, after calculating the gradient magnitude, the result usually needs to be post-processed to ensure that the edge is clear. A common post-processing step is non-maximum suppression (NMS). The purpose of this step is to accurately determine the edge position. It compares the gradient magnitude of each pixel with the neighboring pixel values ​​along the gradient direction. If the gradient magnitude of the current pixel is greater than or equal to the gradient magnitude of the two neighbors along the gradient direction, the pixel is retained as an edge point; otherwise, the pixel is suppressed.

[0059] After non-maximum suppression, the resulting gradient magnitude image usually needs further processing to retain only the obvious edges. A common processing method is edge thresholding. High threshold: determines which pixels belong to obvious edges. Low threshold: determines which pixels are part of the edge, but the edge strength is weak.

[0060] Furthermore, the final result of edge detection is usually based on the combination of these two thresholds: strong edges are retained, and weak edges are also retained if they are connected to strong edges, otherwise they are suppressed. After the above processing, we can get the edge detection result of the image. The edge detection result is usually presented by displaying the edge part in the image as white and the non-edge part as black, such as Figure 3 (b) as shown.

[0061] Furthermore, after obtaining the object contour through edge detection, the grayscale values ​​of the image slices in each laser emission direction within the same object contour are calculated and compared, and the object distance corresponding to the image slice with the highest grayscale value is selected as the real position information of the object.

[0062] like Figure 4 (a) is the image before ghost removal of the point cloud model through edge detection. Figure 4 (b) is the result after ghosting of the point cloud model is eliminated through edge detection.

[0063] The present invention innovatively uses an edge detection algorithm to process FMCW point cloud information, thereby determining the true position of an object. The Sobel operator is innovatively used to implement edge detection of a scanned point image, which has high accuracy and fast detection speed. The relevant formula for edge contour extraction in the FMCW laser point cloud image is given, and the theoretical basis of the present invention is established. The present invention is not only simple in implementation and easy to operate, but also has strong practicality and is suitable for popularization and use.

[0064] The above embodiment is only one of the preferred implementation modes of the present invention and should not be used to limit the protection scope of the present invention. Any changes or modifications that are made to the main design concept and spirit of the present invention and have no substantive significance, and the technical problems they solve are still consistent with the present invention, should be included in the protection scope of the present invention.

Claims

1. A method for removing ghosting from FMCW laser radar point cloud based on edge detection algorithm, characterized in that: It is realized based on FMCW laser radar system; FMCW laser radar system includes tunable laser, optical beam splitter, optical galvanometer, collimating lens, optical coupler, balanced photodetector, data acquisition card and PC; The laser emitted by the tunable laser is divided into two beams by an optical beam splitter, namely, the measuring light and the reference light; the measuring light is deflected by the optical galvanometer, and then collimated by the collimating lens before reaching the object to be measured; the object to be measured generates reflected light, which is coupled with the reference light in the optical coupler after passing through the collimating lens to generate a difference frequency signal; the balanced photodetector performs photoelectric conversion and intensity collection on the difference frequency signal, and the collected difference frequency data is transmitted to the PC through the data acquisition card, and the PC performs frequency resolution based on the collected difference frequency data and its corresponding galvanometer angle to obtain the distance information of the object to be measured; The specific process is: Step 1: The tunable laser performs laser beat frequency by continuously sweeping the frequency-sweeping light source, and the scanning speed of the optical galvanometer matches the frequency-sweeping speed of the light source to obtain a difference frequency signal corresponding to each scanning angle; frequency resolution is performed based on the difference frequency signal to obtain the distance information of the object to be measured; and an object point cloud model is constructed using each scanning angle and the corresponding distance information of the object to be measured to realize lidar scanning; Step 2, extracting the front view of the object point cloud model on a plane perpendicular to the laser emission direction; Step 3: Preprocess the extracted front view, including grayscale image conversion and filtering and smoothing; perform edge detection on the preprocessed image based on the Sobel operator to obtain the object contour; Step 4: Use the object contour to compare the grayscale value of each image slice of the object to be measured in the laser emission direction, and select the object distance corresponding to the image slice with the highest grayscale value as the real position information of the object to be measured, so as to achieve ghost removal.

2. The FMCW laser radar point cloud ghost removal method based on edge detection algorithm according to claim 1 is characterized in that: The wavelength of the tunable laser is λ = 1550nm, the modulation bandwidth is B = 1.09GHz, and the modulation frequency is F m =125KHz, the maximum detection angle is 15 degrees.

3. The FMCW laser radar point cloud ghost removal method based on edge detection algorithm according to claim 1 is characterized in that: In step 3, the grayscale image conversion is expressed as: Y=0.299×R+0.587×G+0.114×B Among them, R, G and B are the pixel values ​​of the red, green and blue channels respectively, and Y is the grayscale value.

4. The FMCW laser radar point cloud ghost removal method based on edge detection algorithm according to claim 1 is characterized in that: In step 3, a Gaussian filter is used to smooth the grayscale converted image.

5. The FMCW laser radar point cloud ghost removal method based on edge detection algorithm according to claim 1 is characterized in that: In step 3, in the Sobel operator, the horizontal gradient operator G used to detect vertical edges x and the vertical gradient operator G for detecting horizontal edges y They are:

6. The FMCW laser radar point cloud ghost removal method based on edge detection algorithm according to claim 5 is characterized in that: In step 3, the Sobel operator is used to calculate the horizontal gradient G of each pixel. x and the vertical gradient G y , then the gradient magnitude G and gradient direction θ of the pixel are:

7. The method for removing ghosting from FMCW laser radar point cloud based on edge detection algorithm according to claim 6, characterized in that: In step 3, the edge position is accurately determined by using non-maximum suppression; if the gradient amplitude of the current pixel is greater than or equal to the gradient amplitudes of the two neighboring pixels along the gradient direction, the pixel is retained as an edge point; Otherwise, suppress the pixel; Edge thresholding is performed, strong edges exceeding the set threshold are retained, weak edges connected to strong edges are retained, and other weak edges are suppressed.

Images