A method and system for unsupervised visual odometry based on texture-free weighted module

Abstract

The present invention relates to an unsupervised visual odometer method and system based on a texture-free weighting module, wherein the method comprises: receiving a monocular video sequence; inputting the monocular video sequence into a prediction model to obtain a prediction result; wherein the prediction model comprises: a depth estimation subnetwork for estimating a depth map corresponding to an input target image; a pose estimation subnetwork for estimating a relative pose between an input target image and a source image; a reprojection module for reprojecting the source image according to the depth map and the relative pose to obtain a reconstructed target image; and a texture-free weighting module for assigning different weights to each pixel point in a texture-free area in a target image and a reconstructed target image to obtain a weighted target image and a weighted reconstructed target image for constructing a reprojection loss. The present invention can reduce the influence of texture-free areas in a scene on the reprojection loss in an unsupervised visual odometer.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to an unsupervised visual odometer method and system based on a texture-free weighted module. Background Art

[0002] As the front end of the real-time localization and mapping (SLAM) system, the visual odometry can estimate the motion of the camera based on the captured images, enabling the mobile robot to achieve self-localization without a preset environmental map, and plays a vital role in the autonomous driving of mobile robots. The earliest methods relied on artificially designed low-dimensional features to obtain the matching point information between two adjacent images, and then calculated the corresponding pose information.

[0003] With the development of deep learning, existing visual odometry can also adopt unsupervised learning. The existing prediction model applied to visual odometry mainly includes a depth estimation subnetwork and a pose estimation subnetwork. The depth estimation subnetwork is used to estimate the depth map corresponding to the image at time t, and the pose estimation subnetwork is used to estimate the relative pose between the image at time t and the image at time t+1. In general, the image at time t is called the target image, and the image at time t+1 is called the source image. According to the above-obtained target image depth map and the relative pose between the target image and the source image, the source image can be reprojected to the target image to obtain the reconstructed target image. The target image and the reconstructed target image should be consistent in theory. According to the difference between the two, a reprojection loss constraint prediction model is constructed for training.

[0004] However, existing prediction models generally rely on reprojection loss. However, when there are textureless areas in the image, the reprojection loss will have large errors. Even if there is no one-to-one correspondence between the target image and the reconstructed target image in the textureless area, a small reprojection loss will still be calculated. This is because each pixel in the textureless area has similar luminosity values, and the corresponding error points still have similar luminosity values. Existing methods determine textureless areas by taking a difference between the target image and the reconstructed target image and directly masking them. Although this alleviates the problem to a certain extent, direct masking will result in the inability to fully utilize image information in the process of calculating the reprojection loss. At the same time, existing methods also have large errors in determining textureless areas. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an unsupervised visual odometer method and system based on a texture-free weighted module, so as to reduce the influence of texture-free areas in a scene on the reprojection loss in the unsupervised visual odometer.

[0006] The technical solution adopted by the present invention to solve the technical problem is: to provide an unsupervised visual odometer method based on a texture-free weighted module, comprising the following steps:

[0007] Receive a monocular video sequence;

[0008] The monocular video sequence is input into a prediction model to obtain the pose between each two adjacent frames in the monocular video sequence and the depth map of each frame image; wherein the prediction model includes:

[0009] The depth estimation subnetwork is used to estimate the depth map corresponding to the input target image;

[0010] The pose estimation subnetwork is used to estimate the relative pose between the input target image and the source image;

[0011] A reprojection module is used to reproject the source image according to the depth map and relative pose to obtain a reconstructed target image;

[0012] The texture-free weighting module is used to assign different weights to each pixel point in the texture-free area in the target image and the reconstructed target image, so as to obtain a weighted target image and a weighted reconstructed target image for constructing a reprojection loss.

[0013] The texture-free weighting module comprises:

[0014] A textureless region screening unit, used for generating a textureless region index of an input image and marking the textureless region in the input image;

[0015] An image filtering unit, used for performing denoising on the texture-free region of the input image according to the texture-free region index to obtain a binary image texture-free region index;

[0016] The weighting unit is used to process the texture-free region index of the binary image using a distance transformation function to obtain a texture-free region weighting weight.

[0017] The texture-free area screening unit comprises:

[0018] A convolution processing subunit, used for performing a convolution operation on an input image to obtain a preliminary index of a texture-free region in the input image;

[0019] The threshold screening processing subunit is used to screen the preliminary index of the texture-free area in the input image through the texture-free threshold to obtain the texture-free area index of the input image.

[0020] When the convolution processing subunit performs a convolution operation, the convolution kernel parameters are set in the following manner:

[0021] When the size of the convolution kernel used is less than or equal to the convolution threshold, all parameters except the convolution kernel center parameter are set to -1, and the convolution kernel center parameter is set to n 2 -1;

[0022] When the size of the convolution kernel used is larger than the convolution threshold, all parameters except the convolution kernel center parameter are set to The convolution kernel center parameter is set to the inverse of the sum of other parameters except the convolution kernel center parameter;

[0023] Among them, n is the size of the convolution kernel, x, y are the position coordinate information of the parameters in the convolution kernel, the origin of the coordinate is at the center of the convolution kernel, and p xy is the parameter of the convolution kernel at the coordinate (x, y), and σ is the preset parameter.

[0024] The technical solution adopted by the present invention to solve the technical problem is: to provide an unsupervised visual odometer system based on a texture-free weighted module, comprising:

[0025] A receiving module, used for receiving a monocular video sequence;

[0026] A prediction module is used to input the monocular video sequence into a prediction model to obtain the pose between each two adjacent frames in the monocular video sequence and the depth map of each frame image; wherein the prediction model includes:

[0027] The depth estimation subnetwork is used to estimate the depth map corresponding to the input target image;

[0028] The pose estimation subnetwork is used to estimate the relative pose between the input target image and the source image;

[0029] A reprojection module is used to reproject the source image according to the depth map and relative pose to obtain a reconstructed target image;

[0030] The texture-free weighting module is used to assign different weights to each pixel point in the texture-free area in the target image and the reconstructed target image, so as to obtain a weighted target image and a weighted reconstructed target image for constructing a reprojection loss.

[0031] The texture-free weighting module comprises:

[0032] A textureless region screening unit, used for generating a textureless region index of an input image and marking the textureless region in the input image;

[0033] An image filtering unit, used for performing denoising on the texture-free region of the input image according to the texture-free region index to obtain a binary image texture-free region index;

[0034] The weighting unit is used to process the texture-free region index of the binary image using a distance transformation function to obtain a texture-free region weighting weight.

[0035] The texture-free area screening unit comprises:

[0036] A convolution processing subunit, used for performing a convolution operation on an input image to obtain a preliminary index of a texture-free region in the input image;

[0037] The threshold screening processing subunit is used to screen the preliminary index of the texture-free area in the input image through the texture-free threshold to obtain the texture-free area index of the input image.

[0038] When the convolution processing subunit performs a convolution operation, the convolution kernel parameters are set in the following manner:

[0039] When the size of the convolution kernel used is less than or equal to the convolution threshold, all parameters except the convolution kernel center parameter are set to -1, and the convolution kernel center parameter is set to n 2 -1;

[0040] When the size of the convolution kernel used is larger than the convolution threshold, all parameters except the convolution kernel center parameter are set to The convolution kernel center parameter is set to the inverse of the sum of other parameters except the convolution kernel center parameter;

[0041] Among them, n is the size of the convolution kernel, x, y are the position coordinate information of the parameters in the convolution kernel, the origin of the coordinate is at the center of the convolution kernel, and p xy is the parameter of the convolution kernel at the coordinate (x, y), and σ is the preset parameter.

[0042] The technical solution adopted by the present invention to solve its technical problem is: to provide an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the unsupervised visual odometer method based on the texture-free weighted module are implemented.

[0043] The technical solution adopted by the present invention to solve its technical problem is: providing a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the unsupervised visual odometer method based on a texture-free weighted module are implemented.

[0044] Beneficial Effects

[0045] Due to the adoption of the above-mentioned technical scheme, the present invention has the following advantages and positive effects compared with the prior art: the present invention introduces a texture-free weighting module, and obtains a weighted target image and a weighted reconstructed target image by respectively inputting the target image and the reconstructed target image into the texture-free weighting module, and then uses the two to construct an improved reprojection loss constraint prediction model for training, thereby effectively reducing the influence of the texture-free area in the scene on the reprojection loss in the unsupervised visual odometer. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of an unsupervised visual odometer method based on a texture-free weighting module according to a first embodiment of the present invention;

[0047] Figure 2 is a schematic diagram of the structure of the prediction model in the first embodiment of the present invention;

[0048] Figure 3 is the original image input in the embodiment;

[0049] Figure 4 A texture-free mask image obtained by using an existing method based on the original image of the embodiment;

[0050] Figure 5 is a schematic diagram of a texture-free region obtained by using the first embodiment of the present invention based on the original image of the embodiment;

[0051] Figure 6 It is a schematic diagram of a texture-free area after denoising obtained by using the first embodiment of the present invention based on the original image of the embodiment. DETAILED DESCRIPTION

[0052] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0053] The first embodiment of the present invention relates to an unsupervised visual odometer method based on a texture-free weighted module, such as Figure 1 As shown, the following steps are included:

[0054] Step 1: Receive a monocular RGB video sequence.

[0055] Step 2: Input the monocular RGB video sequence into the prediction model to obtain the pose between each two adjacent frames in the monocular RGB video sequence and the depth map of each frame image.

[0056] like Figure 2 As shown, the prediction model in this embodiment includes:

[0057] The depth estimation subnetwork is used to estimate the depth map corresponding to the input target image. In this embodiment, the input of the depth estimation subnetwork is the image at time t in the monocular RGB video sequence, that is, the target image, and the output is the depth map corresponding to the image.

[0058] The pose estimation subnetwork is used to estimate the relative pose between the input target image and the source image. In this embodiment, the input of the pose estimation subnetwork is the image at time t and the image at time t+1 in the monocular RGB video sequence, that is, the target image and the source image, and the output is the relative pose between the two images.

[0059] The reprojection module is used to reproject the source image according to the depth map and relative pose to obtain a reconstructed target image.

[0060] The texture-free weighting module is used to assign different weights to each pixel point in the texture-free area in the target image and the reconstructed target image, so as to obtain a weighted target image and a weighted reconstructed target image for constructing a reprojection loss.

[0061] In order to reduce the impact of textureless areas in the scene on the photometric reprojection loss in unsupervised visual odometry, this implementation proposes a textureless weighted module, which aims to assign different weights to each pixel in the textureless area of ​​the image, in order to improve the problem that the reprojection loss of the textureless area in the scene cannot effectively constrain the model for training.

[0062] The textureless weighting module in this embodiment includes a textureless area screening unit, an image filtering unit and a weighting unit connected in sequence. Among them, the goal of the textureless area screening unit is to generate a textureless area index of the input image and mark the textureless area in the input image. In practical applications, the textureless area obtained by screening may have a lot of noise. For this reason, this embodiment adopts an image filtering unit to denoise the textureless area of ​​the input image according to the textureless area index to obtain a binary image textureless area index. Among them, the image filtering unit can adopt a relatively mature Gaussian filtering denoising method. Finally, the weighting unit uses a distance transformation function to process the textureless area index of the binary image to obtain a textureless area weighting weight, and the weighted image can be obtained by multiplying the weight on the image.

[0063] The textureless area screening unit in this embodiment includes: a convolution processing subunit, which is used to perform a convolution operation on the input image to obtain a preliminary index of the textureless area in the input image; and a threshold screening processing subunit, which is used to screen the preliminary index of the textureless area in the input image by a textureless threshold to obtain the textureless area index of the input image. In this embodiment, the preliminary index of the textureless area in the input image is first screened out by convolution screening, and then the preliminary index is further screened by the textureless threshold to obtain the textureless area index of the input image.

[0064] When using convolution screening, assuming that the size of the convolution kernel is n×n, when the convolution kernel size n is small, more detailed texture-free areas will be screened out, but more noise will be generated at the same time; when the convolution kernel size n is large, the image can be effectively smoothed to reduce noise, but only the more macroscopic texture-free areas in the image can be captured, and detailed texture-free areas cannot be screened out well. Therefore, this embodiment designs two methods for setting the convolution kernel parameters, corresponding to the cases of n≤δ and n>δ, respectively, where δ is the convolution threshold. In practical applications, the convolution kernel size n and the appropriate convolution threshold δ can be selected according to different scenarios. In general, the convolution threshold δ is 7. The method for setting the convolution kernel parameters is as follows:

[0065] When the convolution kernel size n≤δ, this implementation sets all parameters except the convolution kernel center parameter to -1, and the convolution kernel center parameter is set to n 2 -1. After the convolution operation is performed on the input image using this convolution kernel, the result obtained in the texture-free area of ​​the input image is close to 0.

[0066] When the convolution kernel size n>δ, the receptive field of the convolution kernel is large. Affected by factors such as lighting and shooting angle, the RGB value corresponding to the edge position of the convolution kernel in the texture-free area of ​​the image and the RGB value corresponding to the center position of the convolution kernel may be quite different. If the above setting method is still used, the error will increase. For this reason, this embodiment proposes a second method for setting the convolution kernel parameters, which can make the parameters of the convolution kernel close to the edge smaller and the parameters close to the center larger. This setting method calculates the other parameters except the convolution kernel center parameter by the following formula, and the convolution kernel center parameter is equal to the negative number of the sum of the other parameters except the convolution kernel center parameter.

[0067]

[0068] Among them, x, y are the position coordinate information of the parameters in the convolution kernel, the origin of the coordinate is at the center of the convolution kernel, and p xy is the parameter of the convolution kernel at the coordinate (x, y), σ is a preset parameter, which can be set according to the actual scenario. The smaller the value, the greater the difference between the parameters of the convolution kernel near the center and the parameters near the edge.

[0069] After the above convolution operation, the preliminary index I of the texture-free area in the image can be obtained. tex , the result obtained in the texture-free area in the image tends to 0. In this implementation, the following formula is used to further screen the texture-free area: Among them, θ is the set threshold, which can be set according to different scenarios. t ' ex is the texture-free area index of the input image, where the area with a value of 1 represents a texture-free area.

[0070] Figure 3 is the original image of the embodiment, Figure 4 is the textureless mask of the existing method, Figure 5 The texture-free area is obtained by screening using the method of this embodiment. Figure 6 is the texture-free area after denoising using the method of this embodiment. Figure 5 and Figure 6 It can be seen from the figure that the texture-free area can be more completely screened out through the processing of the texture-free weighting module in this embodiment. After filtering, Figure 6 The noise in the image is removed to obtain a more accurate texture-free area.

[0071] In this embodiment, both the target image and the reconstructed target image obtained after reprojection are processed by the above-mentioned texture-free weighted module, so that a weighted target image and a weighted reconstructed target image that more accurately reflect the texture-free area can be obtained. The weighted target image and the weighted reconstructed target image are then used to construct a reprojection loss constraint prediction model for training, thereby effectively reducing the impact of the texture-free area in the scene on the reprojection loss in the unsupervised visual odometry.

[0072] A second embodiment of the present invention relates to an unsupervised visual odometer system based on a texture-free weighting module, comprising:

[0073] A receiving module, used for receiving a monocular video sequence;

[0074] A prediction module is used to input the monocular video sequence into a prediction model to obtain the pose between each two adjacent frames in the monocular video sequence and the depth map of each frame image; wherein the prediction model includes:

[0075] The depth estimation subnetwork is used to estimate the depth map corresponding to the input target image;

[0076] The pose estimation subnetwork is used to estimate the relative pose between the input target image and the source image;

[0077] A reprojection module is used to reproject the source image according to the depth map and relative pose to obtain a reconstructed target image;

[0078] The texture-free weighting module is used to assign different weights to each pixel point in the texture-free area in the target image and the reconstructed target image, so as to obtain a weighted target image and a weighted reconstructed target image for constructing a reprojection loss.

[0079] The texture-free weighting module comprises:

[0080] A textureless region screening unit, used for generating a textureless region index of an input image and marking the textureless region in the input image;

[0081] An image filtering unit, used for performing denoising on the texture-free region of the input image according to the texture-free region index to obtain a binary image texture-free region index;

[0082] The weighting unit is used to process the texture-free region index of the binary image using a distance transformation function to obtain a texture-free region weighting weight.

[0083] The texture-free area screening unit comprises:

[0084] A convolution processing subunit, used for performing a convolution operation on an input image to obtain a preliminary index of a texture-free region in the input image;

[0085] The threshold screening processing subunit is used to screen the preliminary index of the texture-free area in the input image through the texture-free threshold to obtain the texture-free area index of the input image.

[0086] When the convolution processing subunit performs a convolution operation, the convolution kernel parameters are set in the following manner:

[0087] When the size of the convolution kernel used is less than or equal to the convolution threshold, all parameters except the convolution kernel center parameter are set to -1, and the convolution kernel center parameter is set to n 2 -1;

[0088] When the size of the convolution kernel used is larger than the convolution threshold, all parameters except the convolution kernel center parameter are set to The convolution kernel center parameter is set to the inverse of the sum of other parameters except the convolution kernel center parameter;

[0089] Among them, n is the size of the convolution kernel, x, y are the position coordinate information of the parameters in the convolution kernel, the origin of the coordinate is at the center of the convolution kernel, and p xy is the parameter of the convolution kernel at the coordinate (x, y), and σ is the preset parameter.

[0090] A third embodiment of the present invention relates to an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the unsupervised visual odometer method based on a texture-free weighted module of the first embodiment when executing the computer program.

[0091] A fourth embodiment of the present invention relates to a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the unsupervised visual odometer method based on a texture-free weighting module of the first embodiment are implemented.

[0092] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) that contain computer-usable program code.

[0093] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0094] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction method, which is implemented in the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0095] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0096] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An unsupervised visual odometry method based on a texture-free weighted module, characterized in that: The following steps are involved: Receive a monocular video sequence; The monocular video sequence is input into a prediction model to obtain the pose between each two adjacent frames in the monocular video sequence and the depth map of each frame image; wherein the prediction model includes: The depth estimation subnetwork is used to estimate the depth map corresponding to the input target image; The pose estimation subnetwork is used to estimate the relative pose between the input target image and the source image; A reprojection module is used to reproject the source image according to the depth map and relative pose to obtain a reconstructed target image; A texture-free weighting module is used to assign different weights to each pixel point in the texture-free area of ​​the target image and the reconstructed target image, so as to obtain a weighted target image and a weighted reconstructed target image for constructing a reprojection loss; the texture-free weighting module comprises: A textureless region screening unit, used for generating a textureless region index of an input image and marking the textureless region in the input image; an image filtering unit, configured to perform denoising on the texture-free region of the input image according to the texture-free region index, Get the texture-free region index of the binary image; The weighting unit is used to process the texture-free region index of the binary image using a distance transformation function to obtain a texture-free region weighting weight.

2. The unsupervised visual odometer method based on a texture-free weighted module according to claim 1, characterized in that: The texture-free area screening unit comprises: A convolution processing subunit, used for performing a convolution operation on an input image to obtain a preliminary index of a texture-free region in the input image; The threshold screening processing subunit is used to screen the preliminary index of the texture-free area in the input image through the texture-free threshold to obtain the texture-free area index of the input image.

3. The unsupervised visual odometer method based on a texture-free weighted module according to claim 2, characterized in that: When the convolution processing subunit performs a convolution operation, the convolution kernel parameters are set in the following manner: When the size of the convolution kernel used is less than or equal to the convolution threshold, all parameters except the convolution kernel center parameter are set to -1, and the convolution kernel center parameter is set to n 2 -1; When the size of the convolution kernel used is larger than the convolution threshold, all parameters except the convolution kernel center parameter are set to The convolution kernel center parameter is set to the inverse of the sum of other parameters except the convolution kernel center parameter; Among them, n is the size of the convolution kernel, x, y are the position coordinate information of the parameters in the convolution kernel, the origin of the coordinate is at the center of the convolution kernel, and p xy is the parameter of the convolution kernel at the coordinate (x, y), and σ is the preset parameter.

4. An unsupervised visual odometry system based on a texture-free weighted module, characterized in that: include: A receiving module, used for receiving a monocular video sequence; A prediction module is used to input the monocular video sequence into a prediction model to obtain the pose between each two adjacent frames in the monocular video sequence and the depth map of each frame image; wherein the prediction model includes: The depth estimation subnetwork is used to estimate the depth map corresponding to the input target image; The pose estimation subnetwork is used to estimate the relative pose between the input target image and the source image; A reprojection module is used to reproject the source image according to the depth map and relative pose to obtain a reconstructed target image; A texture-free weighting module is used to assign different weights to each pixel point in the texture-free area of ​​the target image and the reconstructed target image, so as to obtain a weighted target image and a weighted reconstructed target image for constructing a reprojection loss; the texture-free weighting module comprises: A textureless region screening unit, used for generating a textureless region index of an input image and marking the textureless region in the input image; an image filtering unit, configured to perform denoising on the texture-free region of the input image according to the texture-free region index, Get the texture-free region index of the binary image; The weighting unit is used to process the texture-free region index of the binary image using a distance transformation function to obtain a texture-free region weight.

5. The unsupervised visual odometer system based on texture-free weighted module according to claim 4, characterized in that: The texture-free area screening unit comprises: A convolution processing subunit, used for performing a convolution operation on an input image to obtain a preliminary index of a texture-free region in the input image; The threshold screening processing subunit is used to screen the preliminary index of the texture-free area in the input image through the texture-free threshold to obtain the texture-free area index of the input image.

6. The unsupervised visual odometer system based on the texture-free weighted module according to claim 5, characterized in that: When the convolution processing subunit performs a convolution operation, the convolution kernel parameters are set in the following manner: When the size of the convolution kernel used is less than or equal to the convolution threshold, all parameters except the convolution kernel center parameter are set to -1, and the convolution kernel center parameter is set to n 2 -1; When the size of the convolution kernel used is larger than the convolution threshold, all parameters except the convolution kernel center parameter are set to The convolution kernel center parameter is set to the inverse of the sum of other parameters except the convolution kernel center parameter; Among them, n is the size of the convolution kernel, x, y are the position coordinate information of the parameters in the convolution kernel, the origin of the coordinate is at the center of the convolution kernel, and p xy is the parameter of the convolution kernel at the coordinate (x, y), and σ is the preset parameter.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the unsupervised visual odometer method based on a texture-free weighted module as described in any one of claims 1 to 3 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the unsupervised visual odometer method based on a texture-free weighted module as claimed in any one of claims 1 to 3 are implemented.

Images