A multi-frame vision-based depth measurement method and system

Abstract

The application discloses a kind of depth measurement method and system based on multi-frame vision, which comprises: the field of view range that camera is shot on the reference plane where target detection point is located is obtained, and control calibration plate is located at five positions of field of view range respectively, and the target detection image of calibration plate is collected in each position respectively;Respectively obtain the feature point sequence of each frame target detection image;According to the feature point sequence, the camera yaw angle corresponding to each frame target detection image is calculated based on PnP algorithm;According to camera yaw angle and fixed baseline, the effective baseline corresponding to each frame target detection image is calculated;According to effective baseline and camera yaw angle, the depth value corresponding to each frame target detection image is calculated, and the target depth value is calculated according to the depth value corresponding to each frame target detection image.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of depth measurement, in particular to a depth measurement method based on multi-frame vision and a depth measurement system based on multi-frame vision. BACKGROUND

[0002] Visual depth measurement is a core technology in the field of machine vision, and is widely used in industrial detection, robot navigation, three-dimensional reconstruction and other scenes.

[0003] In related technologies, monocular vision or binocular vision is usually used for depth measurement. However, monocular vision for depth measurement is through perspective transformation and object known size to calculate depth, which is low in cost but poor in precision; binocular vision for depth measurement is to calculate depth by using left and right camera parallax, which is high in precision but depends on hardware calibration and is high in cost. SUMMARY

[0004] To solve the above technical problems, the present application provides a depth measurement method based on multi-frame vision, which can accurately measure depth with a monocular camera, taking into account the cost and precision requirements.

[0005] The technical scheme adopted by the present application is as follows:

[0006] A depth measurement method based on multi-frame vision, comprising the following steps: S1, acquiring the field of view range of a camera at a reference plane where a target detection point is located, and controlling a calibration board to be located at five positions of the field of view range respectively, and acquiring a target detection image of the calibration board at each position respectively; wherein the reference plane is perpendicular to the optical axis of the camera, the five positions of the field of view range include: the center position of the field of view range, the position in the horizontal positive direction of the center position, the position in the horizontal negative direction of the center position, the position in the vertical positive direction of the center position and the position in the vertical negative direction of the center position, and the moving amount of the calibration board at adjacent positions is a fixed baseline; S2, acquiring a feature point sequence of each target detection image respectively; S3, calculating the camera yaw angle corresponding to each target detection image based on PnP algorithm according to the feature point sequence; S4, calculating the effective baseline corresponding to each target detection image according to the camera yaw angle and the fixed baseline; S5, calculating the depth value corresponding to each target detection image according to the effective baseline and the camera yaw angle, and calculating the target depth value according to the depth value corresponding to each target detection image.

[0007] In one embodiment of the present application, the calibration board is a chessboard pattern, and step S2 specifically comprises the following steps: S21, adjusting the light-dark area difference degree of each square of the calibration board in the target detection image respectively to improve the light-dark area difference degree of each square and controlling the increase of the light-dark area difference degree; S22, performing frequency domain analysis on the adjusted image by using a ring band-pass filter in multiple dimensions and angles to obtain invariance feature information of the light-dark area difference change; S23, standardizing the invariance feature information in each dimension and angle and summing the standardized information to obtain a first display image corresponding to the invariance feature information; S24, calculating a feature point measure value for the adjusted image and obtaining a second display image corresponding to the feature point measure value; S25, synthesizing the first display image and the second display image into a feature information synthesis image according to a preset proportion; S26, extracting feature points from the feature information synthesis image according to an adjustable threshold to obtain a first feature point set; and S27, selecting four target feature points corresponding to each square of the calibration board from the first feature point set based on the restriction condition of each square of the calibration board.

[0008] In one embodiment of the present application, the kernel function of the ring band-pass filter is generated by the following formula:

[0009] ,

[0010] wherein r represents the polar coordinate radius of input, represents the polar coordinate angle of input, represents the filter direction, represents the wavelength, represents the radial bandwidth, represents the angle bandwidth.

[0011] In one embodiment of the present application, step S22 specifically comprises the following steps: S221, converting the adjusted image from the spatial domain to the frequency domain by using fast Fourier transform; S222, performing convolution operation on the converted image and the ring band-pass filter in the frequency domain; and S223, extracting the invariance feature information from the convolution operation result.

[0012] In one embodiment of the present application, the effective baseline is calculated by the following formula:

[0013] ,

[0014] wherein, represents the effective baseline corresponding to the current position, represents the camera yaw angle, denotes the fixed baseline.

[0015] In one embodiment of the present application, the depth value corresponding to each frame of the target detection image is calculated by the following formula:

[0016] ,

[0017] wherein, denotes the focal length of the camera at the current position, denotes the parallax of the camera between the current position and the previous position.

[0018] A multi-frame vision-based depth measurement system, comprising: a first acquisition module, configured to acquire a field of view range of a camera shooting on a reference plane where a target detection point is located, and control a calibration board to be located at five positions of the field of view range respectively, and collect a frame of target detection image of the calibration board at each position respectively; wherein the reference plane is perpendicular to the optical axis of the camera, the five positions of the field of view range include: a central position of the field of view range, a position in the horizontal positive direction of the central position, a position in the horizontal negative direction of the central position, a position in the vertical positive direction of the central position and a position in the vertical negative direction of the central position, and the moving amount of the calibration board between adjacent positions is a fixed baseline; a second acquisition module, configured to acquire a feature point sequence of each frame of the target detection image respectively; a first calculation module, configured to calculate a camera yaw angle corresponding to each frame of the target detection image based on a PnP algorithm according to the feature point sequence; a second calculation module, configured to calculate an effective baseline corresponding to each frame of the target detection image according to the camera yaw angle and the fixed baseline; and a third calculation module, configured to calculate a depth value corresponding to each frame of the target detection image according to the effective baseline and the camera yaw angle, and calculate a target depth value according to the depth value corresponding to each frame of the target detection image.

[0019] The beneficial effects of the present application are:

[0020] The present application can accurately perform depth measurement by using a monocular camera, and meets the requirements of cost and precision. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 a flowchart of the multi-frame vision-based depth measurement method of the embodiment of the present application;

[0022] Figure 2 a schematic diagram of the positions of the calibration board of one specific embodiment of the present application;

[0023] Figure 3 a block schematic diagram of the multi-frame vision-based depth measurement system of the embodiment of the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0025] Figure 1 A flowchart of the multi-frame vision-based depth measurement method in the embodiments of the present application.

[0026] As shown in Figure 1 , the multi-frame vision-based depth measurement method in the embodiments of the present application can include the following steps:

[0027] S1, acquiring a field of view range of a camera at a reference plane where a target detection point is located, and controlling a calibration board to be located at five positions of the field of view range respectively, and acquiring a target detection image of the calibration board at each position respectively. Wherein, the reference plane is perpendicular to the optical axis of the camera, and the five positions of the field of view range include: a center position of the field of view range, a position in the horizontal positive direction of the center position, a position in the horizontal negative direction of the center position, a position in the vertical positive direction of the center position, and a position in the vertical negative direction of the center position, and the moving amount of the calibration board between adjacent positions is a fixed baseline.

[0028] Specifically, as shown in Figure 2 , the center position of the field of view range is point A, a plane coordinate system can be established with the center position of the field of view range as the origin, the horizontal direction of the field of view range as the x-axis, and the vertical direction of the field of view range as the y-axis, wherein the position in the horizontal positive direction of the center position is the position in the positive direction of the x-axis, that is, B+; the position in the horizontal negative direction of the center position is the position in the negative direction of the x-axis, that is, B-; the position in the vertical positive direction of the center position is the position in the positive direction of the y-axis, that is, C+; and the position in the vertical negative direction of the center position is the position in the negative direction of the y-axis, that is, C-.

[0029] It should be noted that a space coordinate system can also be established in the direction perpendicular to the x-axis and the y-axis as the z-axis to obtain the space coordinates of the calibration board.

[0030] S2, acquiring a feature point sequence of each target detection image respectively.

[0031] In an embodiment of the present application, the calibration board is an international chessboard pattern, and step S2 specifically includes the following steps:

[0032] S21, respectively, the dynamic adjustment of the light and dark area difference degree of each grid of the target detection image calibration board is carried out to improve the light and dark area difference degree of each grid, and the increase of the light and dark area difference degree is controlled.

[0033] Specifically, the dynamic adjustment of the light and dark area difference degree of each grid is carried out, so that the gray level distribution in each grid is uniform, thereby improving the light and dark area difference degree of each grid, so that the feature points corresponding to each grid can be more highlighted. Further, the increase of the light and dark area difference degree can be controlled to avoid distortion caused by excessive increase, specifically, a threshold can be set, and the pixel number exceeding the threshold is evenly distributed to each gray level.

[0034] S22, for the adjusted image, ring bandpass filter is used for frequency domain analysis in multiple dimensions and angles to obtain invariance feature information for light and dark area difference change.

[0035] In an embodiment of the present application, the kernel function of the ring bandpass filter is generated by the following formula:

[0036]

[0037] Wherein, r represents the input polar coordinate radius, represents the input polar coordinate angle, represents the filter direction, represents the wavelength, represents the radial bandwidth, represents the angle bandwidth.

[0038] Wherein, the parameters in the ring bandpass filter can include: wavelength , control the center frequency characteristic of the filter, set multiple wavelengths by geometric progression (such as 1.2 times) to construct a multi-dimensional filter group; filter direction , by adjusting the filter direction , the texture features of different angles can be captured to construct a multi-angle filter group; radial bandwidth and angle bandwidth , respectively control the scale bandwidth and angle bandwidth, affect the selectivity of the filter. That is, by adjusting the parameters in the ring bandpass filter, the filter has optimal resolution ability in spatial domain and frequency domain at the same time, which can robustly handle the nonlinear radiation difference between images. This characteristic makes the extracted features have strong invariance to light and dark area difference change.

[0039] ​Specifically, in one embodiment of the present application, step S22 specifically comprises the following steps: first, the adjusted image is converted from the spatial domain to the frequency domain by using the fast Fourier transform; then, the converted image is convolved with a ring band-pass filter in the frequency domain; finally, the invariance feature information is extracted from the convolution result.

[0040] S23, the invariance feature information under each dimension and angle is normalized, and the normalized information is summed to obtain a first display image corresponding to the invariance feature information.

[0041] Specifically, first, the invariance feature information under each dimension and angle is normalized to eliminate the dimensional influence and ensure the accuracy of information calculation, and then the normalized information is summed to aggregate the information, thereby obtaining the first display image corresponding to the invariance feature information.

[0042] S24, the feature point measure value is calculated for the adjusted image, and a second display image corresponding to the feature point measure value is obtained.

[0043] Specifically, step S24 specifically comprises the following steps:

[0044] S241, the first-order gradient of the adjusted image in the x-axis direction and the y-axis direction is calculated.

[0045] Specifically, the Sobel operator can be used to calculate the first-order gradient Ix of the adjusted image in the x-axis direction and the first-order gradient Iy in the y-axis direction.

[0046] S242, the weighted sum of the gradient square in the x-axis direction, the weighted sum of the product of the gradients in the x-axis direction and the y-axis direction, and the weighted sum of the gradient square in the y-axis direction are calculated respectively.

[0047] Specifically, the following formula can be used for calculation:

[0048] ,

[0049] wherein, represents the weighted sum of the gradient square in the x-axis direction; represents the weighted sum of the gradient square in the y-axis direction; represents the weighted sum of the product of the gradients in the x-axis direction and the y-axis direction; represents the first-order gradient of the image in the x-axis direction; represents the first-order gradient of the image in the y-axis direction; represents a window function, which can be a Gaussian function.

[0050] S243, constructing a gradient feature matrix according to a weighted sum of squares of gradients in the x-axis direction, a weighted sum of products of gradients in the x-axis direction and the y-axis direction, and a weighted sum of squares of gradients in the y-axis direction.

[0051] wherein the gradient feature matrix can be constructed by the formula

[0052] S244, calculating a feature point measure value according to the gradient feature matrix.

[0053] In an embodiment of the present application, the feature point measure value can be calculated by the following formula:

[0054]

[0055] wherein det() represents a determinant of the gradient feature matrix wherein C represents a constant, preferably, the value range of C is 0.04-0.06;

[0056] S25, synthesizing the first display image and the second display image into a feature information synthesis image according to a preset proportion.

[0057] Specifically, for the first display image and the second display image acquired in different ways, the first display image and the second display image are synthesized into the feature information synthesis image according to a preset proportion, so that the feature information synthesis image has both invariance feature information and gradient highlighting, thereby greatly improving the recognition sensitivity of feature points of different properties.

[0058] S26, performing feature point extraction on the feature information synthesis image according to an adjustable threshold value to acquire a first feature point set.

[0059] Specifically, in an embodiment of the present application, the maximum inter-class variance method can be used to automatically determine an optimal threshold value by maximizing the inter-class variance, select points greater than the optimal threshold value in the feature information synthesis image as the first feature points, and form the first feature point set according to the selected first feature points.

[0060] S27, selecting four target feature points corresponding to each grid in the calibration plate from the first feature point set based on the restriction condition of each grid in the calibration plate.

[0061] ​​​​​​​​Specifically, the restriction condition corresponding to each grid feature point is that the lines connecting each feature point and its adjacent two feature points are perpendicular to each other and have equal distances. Specifically, for the first feature point of each grid, lines are connected with its adjacent two first feature points respectively, and it is judged whether the lines are perpendicular to each other and have equal distances. The first feature points satisfying the restriction condition are selected from the first feature point set as target feature points, so that the pseudo feature points can be effectively removed, and the target feature points can be more accurately identified.

[0062] S3, calculating the camera yaw angle corresponding to each frame of target detection image based on the PnP algorithm according to the feature point sequence.

[0063] Specifically, for the extracted target feature points, the target feature points can be sorted according to the synthesized feature information of each target feature point in the feature information synthesis graph to generate a feature point sequence. Then, the camera yaw angle corresponding to each frame of target detection image is calculated based on the PnP algorithm according to the feature point sequence.

[0064] Wherein, the PnP (Perspective-n-Point, n-point perspective problem) algorithm is an algorithm for estimating the camera pose (including position and direction) from the relationship between multiple two-dimensional points and their corresponding three-dimensional points. Specifically, the PnP algorithm can be used to calculate a rotation vector, and then the rotation vector is converted into a rotation matrix, and the camera yaw angle is extracted through Euler angle decomposition. The specific way of calculating the camera yaw angle corresponding to each frame of target detection image based on the PnP algorithm according to the feature point sequence can be calculated in the manner of the prior art. To avoid redundancy, it will not be described in detail here.

[0065] S4, calculating the effective baseline corresponding to each frame of target detection image according to the camera yaw angle and the fixed baseline.

[0066] In an embodiment of the present application, the effective baseline is calculated by the following formula:

[0067]

[0068] Wherein, represents the effective baseline corresponding to the current position, represents the camera yaw angle, represents the fixed baseline.

[0069] S5, calculating the depth value corresponding to each frame of target detection image according to the effective baseline and the camera yaw angle, and calculating the target depth value according to the depth value corresponding to each frame of target detection image.

[0070] Specifically, in an embodiment of the present application, the depth value corresponding to each frame of target detection image can be calculated by the following formula:

[0071] ​ ,

[0072] in, This indicates the focal length of the camera at the current location. This indicates the parallax between the camera's current position and its previous position.

[0073] Specifically, the depth value corresponding to the target detection image at the center of the field of view can be calculated using the formulas described above. The depth value corresponding to the target detection image at the position along the positive x-axis. The depth value corresponding to the target detection image at the position in the negative x-axis direction. The depth value corresponding to the target detection image at the position in the positive y-axis direction. And the depth value of the target detection image at the position in the negative y-axis direction. .

[0074] Then, the average value of the depth value corresponding to each frame of the target detection image is calculated to obtain the target depth value.

[0075] Therefore, this invention can accurately measure depth using a monocular camera, balancing the requirements of cost and accuracy.

[0076] In summary, the depth measurement method based on multi-frame vision according to embodiments of the present invention acquires the field of view captured by the camera on the reference plane where the target detection point is located, and controls the calibration board to be located at five positions within the field of view, acquiring one frame of target detection image of the calibration board at each position. The reference plane is perpendicular to the optical axis of the camera. The five positions within the field of view include: the center position of the field of view, the position in the horizontal positive direction of the center position, the position in the horizontal negative direction of the center position, the position in the vertical positive direction of the center position, and the position in the vertical negative direction of the center position. The movement of the calibration board between adjacent positions is a fixed baseline. Feature point sequences of each frame of target detection image are acquired. Based on the feature point sequences, the camera yaw angle corresponding to each frame of target detection image is calculated using the PnP algorithm. The effective baseline corresponding to each frame of target detection image is calculated based on the camera yaw angle and the fixed baseline. The depth value corresponding to each frame of target detection image is calculated based on the effective baseline and the camera yaw angle. Finally, the target depth value is calculated based on the depth value corresponding to each frame of target detection image. Therefore, depth measurement can be accurately performed using a single-lens camera, balancing cost and accuracy requirements.

[0077] Corresponding to the depth measurement method based on multi-frame vision in the above embodiments, the present invention also proposes a depth measurement system based on multi-frame vision.

[0078] like Figure 3As shown, the multi-frame vision-based depth measurement system of the embodiment of the present application can include a first acquisition module 100, a second acquisition module 200, a first calculation module 300, a second calculation module 400 and a third calculation module 500.

[0079] The first acquisition module 100 is configured to acquire a field of view range of a camera at a reference plane where a target detection point is located, and control a calibration board to be located at five positions of the field of view range respectively, and acquire a target detection image of the calibration board at each position respectively; the reference plane is perpendicular to an optical axis of the camera, the five positions of the field of view range include a center position of the field of view range, a position in a horizontal positive direction of the center position, a position in a horizontal negative direction of the center position, a position in a vertical positive direction of the center position and a position in a vertical negative direction of the center position, and a moving amount of the calibration board between adjacent positions is a fixed baseline; the second acquisition module 200 is configured to acquire a feature point sequence of each target detection image respectively; the first calculation module 300 is configured to calculate a camera yaw angle corresponding to each target detection image based on a PnP algorithm according to the feature point sequence; the second calculation module 400 is configured to calculate an effective baseline corresponding to each target detection image according to the camera yaw angle and the fixed baseline; and the third calculation module 500 is configured to calculate a depth value corresponding to each target detection image according to the effective baseline and the camera yaw angle, and calculate a target depth value according to the depth value corresponding to each target detection image.

[0080] In an embodiment of the present application, the calibration board is an international chessboard pattern, and the second acquisition module 200 is specifically configured to: dynamically adjust a light-dark area difference degree of each cell of the calibration board in the target detection image respectively, so as to improve the light-dark area difference degree of each cell and control an increase amplitude of the light-dark area difference degree; perform frequency domain analysis on the adjusted image in multiple dimensions and angles respectively by using a ring band-pass filter, so as to acquire invariance feature information for the light-dark area difference change; perform standardization processing on the invariance feature information in each dimension and angle, and perform summation operation on the information after the standardization processing, so as to acquire a first display image corresponding to the invariance feature information; calculate a feature point measure value for the adjusted image, and acquire a second display image corresponding to the feature point measure value; synthesize the first display image and the second display image into a feature information synthesis image according to a preset proportion; extract feature points from the feature information synthesis image according to an adjustable threshold value, so as to acquire a first feature point set; and select four target feature points corresponding to each cell of the calibration board from the first feature point set based on a restriction condition of each cell of the calibration board.

[0081] In an embodiment of the present application, a kernel function of the ring band-pass filter is generated by the following formula:

[0082] ,

[0083] wherein r represents an input polar coordinate radius, represents an input polar coordinate angle, represents a filter direction, represents a wavelength, represents a radial bandwidth, represents an angle bandwidth.

[0084] In an embodiment of the present application, the second acquisition module 200 is specifically configured to: convert the adjusted image from a spatial domain to a frequency domain by using a fast Fourier transform; perform a convolution operation on the converted image and an annular band-pass filter in the frequency domain; and extract invariance feature information from a convolution operation result.

[0085] In an embodiment of the present application, the effective baseline is calculated by the following formula:

[0086]

[0087] wherein, represents an effective baseline corresponding to a current position, represents a camera yaw angle, represents a fixed baseline.

[0088] In an embodiment of the present application, the depth value corresponding to each frame of target detection image is calculated by the following formula:

[0089]

[0090] wherein, represents a focal length of a camera at a current position, represents a parallax of the camera between the current position and a previous position.

[0091] It should be noted that details not disclosed in the depth measurement system based on multiple frames of vision in the embodiments of the present application are for reference to the details disclosed in the depth measurement method based on multiple frames of vision described above, and will not be described in detail here.

[0092] ​​According to the multi-frame vision-based depth measurement system provided by the embodiment of the present application, the first acquisition module acquires the field of view range of the camera shooting on the reference plane where the target detection point is located, and controls the calibration board to be located at five positions of the field of view range respectively, and acquires a target detection image of the calibration board at each position respectively; the reference plane is perpendicular to the optical axis of the camera, the five positions of the field of view range include: the central position of the field of view range, the position in the horizontal positive direction of the central position, the position in the horizontal reverse direction of the central position, the position in the vertical positive direction of the central position and the position in the vertical reverse direction of the central position, and the moving amount of the calibration board between adjacent positions is a fixed baseline; the second acquisition module acquires the feature point sequence of each target detection image respectively; the first calculation module calculates the camera yaw angle corresponding to each target detection image based on the PnP algorithm according to the feature point sequence; the second calculation module calculates the effective baseline corresponding to each target detection image according to the camera yaw angle and the fixed baseline; the third calculation module calculates the depth value corresponding to each target detection image according to the effective baseline and the camera yaw angle, and calculates the target depth value according to the depth value corresponding to each target detection image. Thus, the monocular camera can be used to accurately measure the depth, and the cost and accuracy requirements are considered.

[0093] In the description of the present application, the terms "first", "second" are only used for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. The meaning of "plurality" is two or more, unless otherwise specifically limited.

[0094] In the present application, unless otherwise specifically defined and limited, the terms "mounting", "connection", "connecting", "fixing" and the like should be understood in a broad sense, for example, can be fixed connection, can also be detachable connection, or integral; can be mechanical connection, can also be electrical connection; can be directly connected, can also be indirectly connected through an intermediate medium, can be the internal communication of two elements or the interaction relationship between two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0095] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0096] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0097] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0098] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A depth measurement method based on multi-frame vision, characterized in that, Includes the following steps: S1, acquire the field of view captured by the camera on the reference plane where the target detection point is located, and control the calibration board to be located at five positions within the field of view, and acquire one frame of the target detection image of the calibration board at each position; wherein, the reference plane is perpendicular to the optical axis of the camera, and the five positions of the field of view include: the center position of the field of view, the position in the horizontal positive direction of the center position, the position in the horizontal negative direction of the center position, the position in the vertical positive direction of the center position, and the position in the vertical negative direction of the center position, and the amount of movement of the calibration board in adjacent positions is a fixed baseline; S2, acquire the feature point sequence of each frame of the target detection image; wherein, the calibration board is a chessboard pattern, and step S2 specifically includes the following steps: S21, dynamically adjust the degree of difference between light and dark areas for each grid of the calibration plate in the target detection image to improve the degree of difference between light and dark areas for each grid, and control the increase in the degree of difference between light and dark areas; S22, For the adjusted image, frequency domain analysis is performed using a ring bandpass filter in multiple dimensions and angles to obtain invariant feature information for changes in brightness and darkness. S23, standardize the invariant feature information under each dimension and angle, and sum the standardized information to obtain the first display image corresponding to the invariant feature information; S24, calculate the feature point measure value for the adjusted image, and obtain the second display image corresponding to the feature point measure value; S25, the first display image and the second display image are combined into a composite image with feature information according to a preset weight; S26, extract feature points from the feature information synthesis map according to the adjustable threshold value to obtain a first feature point set; S27, Based on the constraints of each grid in the calibration board, select four target feature points corresponding to each grid in the calibration board from the first feature point set; S3, calculate the camera yaw angle corresponding to each frame of the target detection image based on the PnP algorithm according to the feature point sequence; S4, calculate the effective baseline corresponding to each frame of the target detection image based on the camera yaw angle and the fixed baseline; S5, calculate the depth value corresponding to each frame of the target detection image based on the effective baseline and the camera yaw angle, and calculate the target depth value based on the depth value corresponding to each frame of the target detection image.

2. The depth measurement method based on multi-frame vision according to claim 1, characterized in that, The kernel function of the ring bandpass filter is generated using the following formula: ， Where r represents the input polar coordinate radius, This represents the input polar coordinate angle. Indicates the direction of the filter. Indicates wavelength. Indicates radial bandwidth. Indicates angular bandwidth.

3. The depth measurement method based on multi-frame vision according to claim 2, characterized in that, Step S22 specifically includes the following steps: S221, uses Fast Fourier Transform to transform the adjusted image from the spatial domain to the frequency domain; S222, In the frequency domain, the converted image is convolved with the ring bandpass filter; S223, extract the invariant feature information from the convolution operation result.

4. The depth measurement method based on multi-frame vision according to claim 1, characterized in that, The effective baseline is calculated using the following formula: ， in, This indicates the effective baseline corresponding to the current location. This indicates the camera's yaw angle. This refers to the fixed baseline.

5. The depth measurement method based on multi-frame vision according to claim 1, characterized in that, The depth value corresponding to each frame of the target detection image is calculated using the following formula: ， in, This indicates the focal length of the camera at the current location. This indicates the parallax between the camera's current position and its previous position.

6. A depth measurement system based on multi-frame vision, characterized in that, include: The first acquisition module is used to acquire the field of view captured by the camera on the reference plane where the target detection point is located, and to control the calibration plate to be located at five positions within the field of view, and to acquire a target detection image of the calibration plate at each position; wherein, the reference plane is perpendicular to the optical axis of the camera, and the five positions within the field of view include: the center position of the field of view, the position in the horizontal positive direction of the center position, the position in the horizontal negative direction of the center position, the position in the vertical positive direction of the center position, and the position in the vertical negative direction of the center position, and the amount of movement of the calibration plate between adjacent positions is a fixed baseline; The second acquisition module is used to acquire feature point sequences for each frame of the target detection image. The calibration board is a chessboard pattern. Specifically, the second acquisition module is used to: dynamically adjust the degree of difference between bright and dark areas in each square of the calibration board in the target detection image to increase the degree of difference between bright and dark areas in each square, and control the increase in the degree of difference between bright and dark areas; for the adjusted image, perform frequency domain analysis using a ring bandpass filter in multiple dimensions and angles to obtain invariant feature information to changes in the difference between bright and dark areas; and analyze the invariant feature information in each dimension and angle. The variable feature information is standardized, and the standardized information is summed to obtain a first display image corresponding to the invariant feature information; feature point measure values ​​are calculated for the adjusted image, and a second display image corresponding to the feature point measure values ​​is obtained; the first display image and the second display image are combined into a feature information composite image according to a preset weight; feature points are extracted from the feature information composite image according to an adjustable threshold value to obtain a first feature point set; based on the constraints of each grid in the calibration board, four target feature points corresponding to each grid in the calibration board are selected from the first feature point set; A first calculation module is used to calculate the camera yaw angle corresponding to each frame of the target detection image based on the PnP algorithm according to the feature point sequence. The second calculation module is used to calculate the effective baseline corresponding to each frame of the target detection image based on the camera yaw angle and the fixed baseline. The third calculation module is used to calculate the depth value corresponding to each frame of the target detection image based on the effective baseline and the camera yaw angle, and to calculate the target depth value based on the depth value corresponding to each frame of the target detection image.

Images