Track extraction method and system based on pure vision, medium and equipment
Through the pure vision-based orbit extraction method, semantic segmentation and template matching algorithms are used to solve the problem of extraction difficulties under expensive and complex road conditions of lidar equipment, and the track information extraction in efficient operation of steel plants is realized, improving robustness and practical value.
Patent Information
- Application Number
- CN202410921935.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-10
- Publication Date
- 2024-11-12
AI Technical Summary
The existing track extraction technology relies on expensive lidar equipment, and it is difficult to achieve real-time and accurate track extraction in areas with low altitude differences and complex road conditions, which cannot meet the needs of efficient operation of steel plants.
The track extraction method based on pure vision is adopted to obtain details and semantic features through the semantic segmentation model, combine template matching algorithms and view angle conversion to extract the track position, and use similarity comparison to determine the track position to achieve the track information extraction of the current driving direction.
It improves the robustness and practical value of track extraction, can accurately extract track information of the bicycle driving area under various road conditions, supports railway safety analysis and related tasks, and reduces equipment costs.
Smart Images

Figure CN118941764A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of template matching and visual conversion, and in particular to a track extraction method, system, medium and device based on pure vision. Background Art
[0002] The efficient operation of a steel plant is inseparable from the precise material handling and distribution, in which the track system plays a core role. This system is not only responsible for the transportation of raw materials, semi-finished products and finished products, but also plays an irreplaceable role in all aspects of production. At present, many track extraction solutions use LiDAR equipment to extract tracks based on the height difference between the rails and sleepers. However, such equipment is expensive and has limitations. For areas with insignificant height differences, such as gate areas, the location of the rails cannot be inferred based on the height difference.
[0003] The prior art (Chinese patent application number CN201410544556.7) discloses a railway automatic measurement method based on point cloud data. This method uses a geometric fitting method combined with a laser radar device, but does not consider the real-time performance of track extraction and the extraction of the vehicle track driving area under complex road conditions. In addition, the laser radar device is expensive.
[0004] Patent document CN115457379A discloses a method and system for extracting roads from remote sensing images by combining semantic segmentation and angle prediction, which calculates road angle information from real road information, explicitly reflects the geometric feature that the road inclination angle has a certain stability, and provides the road angle true value for comparison by adding constraints to the geometric feature for the neural network. However, this patent cannot completely solve the existing technical problems, nor can it meet the needs of the present invention. Summary of the invention
[0005] In view of the defects in the prior art, the object of the present invention is to provide a track extraction method, system, medium and equipment based on pure vision.
[0006] The pure vision-based track extraction method provided by the present invention includes:
[0007] Step 1: Input the image into the open-source bilateral segmentation model based on semantics, obtain detail features and semantic features through the detail branch and semantic branch respectively, then fuse the two features through the feature fusion module, and accelerate the model training through the auxiliary training acceleration strategy, and finally output the segmentation result to obtain the mask map of the track;
[0008] Step 2: According to the current positioning, obtain the 3D position coordinates of the task point ahead in the map from the predefined waypoint map, use the calibrated camera extrinsic matrix to transform the 3D task point coordinates into the camera coordinate system, then use the camera intrinsic matrix to project the 3D point in the camera coordinate system onto the 2D image plane, and normalize it to obtain the final image pixel coordinates. Finally, generate a template of preset width as the standard based on the transformed task point;
[0009] Step 3: Convert the mask in step 1 and the template in step 2 to a top view perspective;
[0010] Step 4: Rotate the template within a preset angle range, find the best matching result in the mask image through template matching, and determine the position of one of the rails;
[0011] Step 5: Calculate the similarity between the mask image and the template of the same size at the pixels within the preset range on the left and right sides of the determined rail, and identify the side with the highest similarity as the position of the other rail.
[0012] Preferably, step 2 comprises:
[0013] The camera extrinsic matrix transforms the three-dimensional task point Convert to camera coordinate system The expression is:
[0014]
[0015] Among them, R is a 3×3 rotation matrix; T is a 3×1 translation vector;
[0016] The camera intrinsic parameter matrix converts the three-dimensional coordinates in the camera coordinate system Convert to a two-dimensional image The expression is:
[0017]
[0018] Among them, f x ,f y is the focal length in the x and y directions; x0, y0 are the coordinate positions of the principal point.
[0019] Preferably, the step 3 comprises:
[0020] The perspective change matrix is used for perspective transformation. The perspective change matrix describes the change relationship between the original image pixel point (x, y) and the new pixel point (x′, y′) after the change. Specifically:
[0021]
[0022] in: Belongs to the transformation matrix, including scaling and rotation; Belongs to the translation vector; [c1 c2] belongs to the projection vector. For affine transformation, the projection vector is equal to 0; Where t represents the normalization factor in the homogeneous coordinates, and the final two-dimensional coordinates (x′, y′) are obtained after normalization.
[0023] Preferably, step 4 comprises:
[0024] Template matching uses mean shift cross correlation as the correlation criterion. The algorithm first subtracts the average value of the entire image and then performs cross-correlation calculation. The specific formula is as follows:
[0025]
[0026] in:
[0027] T′(x′,y′)=T(x′,y′)-1 / (w·h)·∑ x″,y″ T(x″,y″)
[0028] I′(x+x′,y+y′)
[0029] =I(x+x′,y+y′)-1 / (w·h)·∑ x″,y″ I(x+x″.y+y″) 2
[0030] Where I represents the target image, T represents the template, and R represents the final result; I′ represents the result of subtracting the corresponding average value at each pixel position of the target image, and T′ represents the result of subtracting the corresponding average value at each pixel position of the template; x, y represent the horizontal and vertical coordinates of the target image pixels; W, H represent the width and length of the image; x′, x″ represent the horizontal coordinates of the template pixels; y′, y″ represent the vertical coordinates of the template pixels; w, h represent the width and length of the template; where x∈[0, Ww], y∈[0, Hh], x′, x″∈[0, w-1], y′, y″∈[0, h-1].
[0031] The pure vision-based track extraction system provided by the present invention comprises:
[0032] Module M1: Input the image into the open source model of bilateral segmentation based on semantics, obtain detail features and semantic features through detail branch and semantic branch respectively, then fuse the two features through feature fusion module, accelerate model training through auxiliary training acceleration strategy, and finally output the segmentation result to obtain the mask map of the track;
[0033] Module M2: According to the current positioning, the 3D position coordinates of the task point ahead are obtained from the predefined waypoint map, and the 3D task point coordinates are converted to the camera coordinate system using the calibrated camera extrinsic matrix. The 3D points in the camera coordinate system are then projected onto the 2D image plane using the camera intrinsic matrix, and the final image pixel coordinates are normalized. Finally, a template of preset width is generated as a standard based on the converted task point.
[0034] Module M3: Convert the mask in module M1 and the template in module M2 to a top view perspective;
[0035] Module M4: Rotate the template within a preset angle range, find the best matching result in the mask image through template matching, and determine the position of one of the rails;
[0036] Module M5: Calculate the similarity between the mask image and the template of the same size at the pixels within the preset range on the left and right sides of the determined rail, and identify the side with the highest similarity as the position of the other rail.
[0037] Preferably, the module M2 comprises:
[0038] The camera extrinsic matrix transforms the three-dimensional task point Convert to camera coordinate system The expression is:
[0039]
[0040] Among them, R is a 3×3 rotation matrix; T is a 3×1 translation vector;
[0041] The camera intrinsic parameter matrix converts the three-dimensional coordinates in the camera coordinate system Convert to a two-dimensional image The expression is:
[0042]
[0043] Among them, f x ,f y is the focal length in the x and y directions; x0, y0 are the coordinate positions of the principal point.
[0044] Preferably, the module M3 comprises:
[0045] The perspective change matrix is used for perspective transformation. The perspective change matrix describes the change relationship between the original image pixel point (x, y) and the new pixel point (x′, y′) after the change. Specifically:
[0046]
[0047] in: Belongs to the transformation matrix, including scaling and rotation; Belongs to the translation vector; [c1 c2] belongs to the projection vector. For affine transformation, the projection vector is equal to 0; Where t represents the normalization factor in the homogeneous coordinates, and the final two-dimensional coordinates (x′, y′) are obtained after normalization.
[0048] Preferably, the module M4 comprises:
[0049] Template matching uses mean shift cross correlation as the correlation criterion. The algorithm first subtracts the average value of the entire image and then performs cross-correlation calculation. The specific formula is as follows:
[0050]
[0051] in:
[0052] T′(x′,y′)=T(x′,y′)-1 / (w·h)·∑ x″,y″ T(x″,y″)
[0053] I′(x+x′,y+y′)
[0054] =I(x+x′,y+y′)-1 / (w·h)·∑ x″,y″ I(x+x″.y+y″) 2
[0055] Where I represents the target image, T represents the template, and R represents the final result; I′ represents the result of subtracting the corresponding average value at each pixel position of the target image, and T′ represents the result of subtracting the corresponding average value at each pixel position of the template; x, y represent the horizontal and vertical coordinates of the target image pixels; W, H represent the width and length of the image; x′, x″ represent the horizontal coordinates of the template pixels; y′, y″ represent the vertical coordinates of the template pixels; w, h represent the width and length of the template; where x∈[0, Ww], y∈[0, Hh], x′, x″∈[0, w-1], y′, y″∈[0, h-1].
[0056] According to the computer-readable storage medium storing a computer program provided by the present invention, when the computer program is executed by a processor, the steps of the pure vision-based track extraction method are implemented.
[0057] The electronic device provided according to the present invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the computer program implements the steps of the pure vision-based track extraction method when executed by the processor.
[0058] Compared with the prior art, the present invention has the following beneficial effects:
[0059] (1) The present invention can remove other interfering tracks other than the driving direction of the vehicle from the environment and only retain track information that is meaningful to the current driving direction, thus having high robustness and practical value;
[0060] (2) The present invention provides a method for extracting the vehicle's travel track area under various road conditions based on pure vision. By combining the use of semantic segmentation models, template matching algorithms, perspective conversion, and similarity comparison, the track of the vehicle wash travel area can be accurately extracted, providing sufficient space and data support for subsequent railway safety analysis and various related tasks. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:
[0062] Figure 1 Flowchart of the pure vision-based track extraction method. DETAILED DESCRIPTION
[0063] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.
[0064] Example 1
[0065] The present invention provides a track extraction method based on pure vision, such as Figure 1 , extract the track in the direction of the vehicle's travel in the image according to the following steps;
[0066] Step 1: Input the image into the semantic-based bilateral segmentation open source model BiSeNet V2, obtain detail features and semantic features through the detail branch and semantic branch respectively, then fuse the two features through the feature fusion module, and accelerate the model training through the auxiliary training acceleration strategy, and finally output the segmentation result to obtain the track mask;
[0067] Step 2: According to the current positioning, obtain the 3D position coordinates of the task point ahead in the map from the predefined waypoint map, use the calibrated camera extrinsic matrix to transform the 3D task point coordinates into the camera coordinate system, then use the camera intrinsic matrix to project the 3D point in the camera coordinate system onto the 2D image plane, and normalize it to obtain the final image pixel coordinates. Finally, generate a template of preset width as the standard based on the transformed task point;
[0068] Step 3: Convert the mask in step 1 and the template in step 2 to a top view perspective;
[0069] Step 4: Rotate the template within a preset angle range, find the best matching result in the mask image through template matching, and determine the position of one of the rails;
[0070] Step 5: Calculate the similarity between the mask image and the template of the same size at the pixels within the preset range on the left and right sides of the determined rail, and identify the side with the highest similarity as the position of the other rail.
[0071] The step 2 comprises:
[0072] The camera extrinsic matrix transforms the three-dimensional task point Convert to camera coordinate system The expression is:
[0073]
[0074] Among them, R is a 3×3 rotation matrix; T is a 3×1 translation vector;
[0075] The camera intrinsic parameter matrix converts the three-dimensional coordinates in the camera coordinate system Convert to a two-dimensional image The expression is:
[0076]
[0077] Among them, f x ,f y is the focal length in the x and y directions; x0, y0 are the coordinate positions of the principal point.
[0078] The step 3 comprises:
[0079] The perspective change matrix is used for perspective transformation. The perspective change matrix describes the change relationship between the original image pixel point (x, y) and the new pixel point (x′, y′) after the change. Specifically:
[0080]
[0081] in: Belongs to the transformation matrix, including scaling and rotation; Belongs to the translation vector; [c1 c2] belongs to the projection vector. For affine transformation, the projection vector is equal to 0; Where t represents the normalization factor in the homogeneous coordinates, and the final two-dimensional coordinates (x′, y′) are obtained after normalization.
[0082] The step 4 comprises:
[0083] Template matching uses mean shift cross correlation (TM_CCOEFF) as the correlation criterion. The algorithm first subtracts the average value of the entire image and then performs cross-correlation calculation. The specific formula is as follows:
[0084]
[0085] in:
[0086] T′(x′,y′)=T(x′,y′)-1 / (w·h)·Σ x″,y″ T(x″,y″)
[0087] I′(x+x′,y+y′)
[0088] =I(x+x′,y+y′)-1 / (w·h)·∑ x″,y″ I(x+x″.y+y″) 2
[0089] Where I represents the target image, T represents the template, and R represents the final result; I′ represents the result of subtracting the corresponding average value from each pixel position of the target image, and T′ represents the result of subtracting the corresponding average value from each pixel position of the template; x, y represent the horizontal and vertical coordinates of the target image pixels; W, H represent the width and length of the image; x', x" represent the horizontal coordinates of the template pixels; y', y" represent the vertical coordinates of the template pixels; w, h represent the width and length of the template;
[0090] in:
[0091] x∈[0,Ww]
[0092] y∈[0,Hh]
[0093] x′,x″∈[0,w-1]
[0094] y′,y″∈[0,h-1]
[0095] This algorithm can reduce the interference of outliers and has a small amount of calculation, making it suitable for real-time tracking scenarios.
[0096] Example 2
[0097] The present invention also provides a pure vision-based track extraction system, which can be implemented by executing the process steps of the pure vision-based track extraction method, that is, those skilled in the art can understand the pure vision-based track extraction method as a preferred implementation of the pure vision-based track extraction system.
[0098] The pure vision-based track extraction system provided by the present invention includes: module M1: inputting an image into a semantic-based bilateral segmentation open source model, obtaining detail features and semantic features through detail branches and semantic branches respectively, and then fusing the two features through a feature fusion module, and accelerating model training through an auxiliary training acceleration strategy, and finally outputting a segmentation result to obtain a track mask; module M2: obtaining the three-dimensional position coordinates of the front task point in the map from a predefined waypoint map according to the current positioning, using a calibrated camera extrinsic matrix to convert the three-dimensional task point coordinates into a camera coordinate system, and then using a camera intrinsic matrix to convert the camera coordinates into a camera coordinate system. The three-dimensional points in the reference system are projected onto the two-dimensional image plane and normalized to obtain the final image pixel coordinates. Finally, a template of preset width is generated as a standard based on the converted task points. Module M3: The mask in module M1 and the template in module M2 are converted to a top-down view. Module M4: The template is rotated within a preset angle range, and the optimal matching result in the mask is found through template matching to determine the position of one of the rails. Module M5: The similarity between the mask and the template of the same size is calculated at the pixels within a preset range on the left and right sides of the determined rail, and the side with the highest similarity is identified as the position of the other rail.
[0099] The module M2 comprises:
[0100] The camera extrinsic matrix transforms the three-dimensional task point Convert to camera coordinate system The expression is:
[0101]
[0102] Among them, R is a 3×3 rotation matrix; T is a 3×1 translation vector;
[0103] The camera intrinsic parameter matrix converts the three-dimensional coordinates in the camera coordinate system Convert to a two-dimensional image The expression is:
[0104]
[0105] Among them, f x ,f y is the focal length in the x and y directions; x0, y0 are the coordinate positions of the principal point.
[0106] The module M3 comprises:
[0107] The perspective change matrix is used for perspective transformation. The perspective change matrix describes the change relationship between the original image pixel point (x, y) and the new pixel point (x′, y′) after the change. Specifically:
[0108]
[0109] in: Belongs to the transformation matrix, including scaling and rotation; Belongs to the translation vector; [c1 c2] belongs to the projection vector. For affine transformation, the projection vector is equal to 0; Where t represents the normalization factor in the homogeneous coordinates, and the final two-dimensional coordinates (x′, y′) are obtained after normalization.
[0110] The module M4 comprises:
[0111] Template matching uses mean shift cross correlation as the correlation criterion. The algorithm first subtracts the average value of the entire image and then performs cross-correlation calculation. The specific formula is as follows:
[0112]
[0113] in:
[0114] T′(x′,y′)=T(x′,y′)-1 / (w·h)·∑ x″,y″ T(x″,y″)
[0115] I′(x+x′,y+y′)
[0116] =I(x+x′,y+y′)-1 / (w·h)·∑ x″,y″ I(x+x″.y+y″) 2
[0117] Where I represents the target image, T represents the template, and R represents the final result; I′ represents the result of subtracting the corresponding average value at each pixel position of the target image, and T′ represents the result of subtracting the corresponding average value at each pixel position of the template; x, y represent the horizontal and vertical coordinates of the target image pixels; W, H represent the width and length of the image; x′, x″ represent the horizontal coordinates of the template pixels; y′, y″ represent the vertical coordinates of the template pixels; w, h represent the width and length of the template; where x∈[0, Ww], y∈[0, Hh], x′, x″∈[0, w-1], y′, y″∈[0, h-1].
[0118] Those skilled in the art know that, in addition to implementing the system, device and its various modules provided by the present invention in a purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, the system, device and its various modules provided by the present invention can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing the method and structures within the hardware component.
[0119] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.
Claims
1. A track extraction method based on pure vision, characterized in that: include: Step 1: Input the image into the open-source bilateral segmentation model based on semantics, obtain detail features and semantic features through the detail branch and semantic branch respectively, then fuse the two features through the feature fusion module, and accelerate the model training through the auxiliary training acceleration strategy, and finally output the segmentation result to obtain the mask map of the track; Step 2: According to the current positioning, obtain the 3D position coordinates of the task point ahead in the map from the predefined waypoint map, use the calibrated camera extrinsic matrix to transform the 3D task point coordinates into the camera coordinate system, then use the camera intrinsic matrix to project the 3D point in the camera coordinate system onto the 2D image plane, and normalize it to obtain the final image pixel coordinates. Finally, generate a template of preset width as the standard based on the transformed task point; Step 3: Convert the mask in step 1 and the template in step 2 to a top view perspective; Step 4: Rotate the template within a preset angle range, find the best matching result in the mask image through template matching, and determine the position of one of the rails; Step 5: Calculate the similarity between the mask image and the template of the same size at the pixels within the preset range on the left and right sides of the determined rail, and identify the side with the highest similarity as the position of the other rail.
2. The pure vision-based track extraction method according to claim 1, characterized in that: The step 2 comprises: The camera extrinsic matrix transforms the three-dimensional task point Convert to camera coordinate system The expression is: Among them, R is a 3×3 rotation matrix; T is a 3×1 translation vector; The camera intrinsic parameter matrix converts the three-dimensional coordinates in the camera coordinate system Convert to a two-dimensional image The expression is: Among them, f x ,f y is the focal length in the x and y directions; x0, y0 are the coordinate positions of the principal point.
3. The pure vision-based track extraction method according to claim 1, characterized in that: The step 3 comprises: The perspective change matrix is used for perspective transformation. The perspective change matrix describes the change relationship between the original image pixel point (x, y) and the new pixel point (x′, y′) after the change. Specifically: in: Belongs to the transformation matrix, including scaling and rotation; Belongs to the translation vector; [c1 c2] belongs to the projection vector. For affine transformation, the projection vector is equal to 0; Where t represents the normalization factor in the homogeneous coordinates, and the final two-dimensional coordinates (x′, y′) are obtained after normalization.
4. The pure vision-based track extraction method according to claim 1, characterized in that: The step 4 comprises: Template matching uses mean shift cross correlation as the correlation criterion. The algorithm first subtracts the average value of the entire image and then performs cross-correlation calculation. The specific formula is as follows: in: T′(x′,y′)=T(x′,y′)-1 / (w·h)·∑ x″,y″ T(x″,y″) I′(x+x′,y+y′) =I(x+x′,y+y′)-1 / (w·h)·∑ x″,y″ I(x+x″.y+y″) 2 Where I represents the target image, T represents the template, and R represents the final result; I′ represents the result of subtracting the corresponding average value at each pixel position of the target image, and T′ represents the result of subtracting the corresponding average value at each pixel position of the template; x, y represent the horizontal and vertical coordinates of the target image pixels; W, H represent the width and length of the image; x′, x″ represent the horizontal coordinates of the template pixels; y′, y″ represent the vertical coordinates of the template pixels; w, h represent the width and length of the template; where x∈[0, Ww], y∈[0, Hh], x′, x″∈[0, w-1], y′, y″∈[0, h-1].
5. A pure vision-based track extraction system, characterized in that: include: Module M1: Input the image into the open source model of bilateral segmentation based on semantics, obtain detail features and semantic features through detail branch and semantic branch respectively, then fuse the two features through feature fusion module, accelerate model training through auxiliary training acceleration strategy, and finally output the segmentation result to obtain the mask map of the track; Module M2: According to the current positioning, the 3D position coordinates of the task point ahead are obtained from the predefined waypoint map, and the 3D task point coordinates are converted to the camera coordinate system using the calibrated camera extrinsic matrix. The 3D points in the camera coordinate system are then projected onto the 2D image plane using the camera intrinsic matrix, and the final image pixel coordinates are normalized. Finally, a template of preset width is generated as a standard based on the converted task point. Module M3: Convert the mask in module M1 and the template in module M2 to a top view perspective; Module M4: Rotate the template within a preset angle range, find the best matching result in the mask image through template matching, and determine the position of one of the rails; Module M5: Calculate the similarity between the mask image and the template of the same size at the pixels within the preset range on the left and right sides of the determined rail, and identify the side with the highest similarity as the position of the other rail.
6. The pure vision-based track extraction system according to claim 5, characterized in that: The module M2 comprises: The camera extrinsic matrix transforms the three-dimensional task point Convert to camera coordinate system The expression is: Among them, R is a 3×3 rotation matrix; T is a 3×1 translation vector; The camera intrinsic parameter matrix converts the three-dimensional coordinates in the camera coordinate system Convert to a two-dimensional image The expression is: Among them, f x ,f y is the focal length in the x and y directions; x0, y0 are the coordinate positions of the principal point.
7. The pure vision-based track extraction system according to claim 5, characterized in that: The module M3 comprises: The perspective change matrix is used for perspective transformation. The perspective change matrix describes the change relationship between the original image pixel point (x, y) and the new pixel point (x′, y′) after the change. Specifically: in: Belongs to the transformation matrix, including scaling and rotation; Belongs to the translation vector; [c1 c2] belongs to the projection vector. For affine transformation, the projection vector is equal to 0; Where t represents the normalization factor in the homogeneous coordinates, and the final two-dimensional coordinates (x′, y′) are obtained after normalization.
8. The purely visual track extraction system according to claim 5, characterized in that: The module M4 comprises: Template matching uses mean shift cross correlation as the correlation criterion. The algorithm first subtracts the average value of the entire image and then performs cross-correlation calculation. The specific formula is as follows: in: T′(x′,y′)=T(x′,y′)-1 / (w·h)·∑ x″,y″ T(x″,y″) I′(x+x′,y+y′) =I(x+x′,y+y′)-1 / (w·h)·Σ x″,y″ I(x+x″.y+y″) 2 Where I represents the target image, T represents the template, and R represents the final result; I′ represents the result of subtracting the corresponding average value at each pixel position of the target image, and T′ represents the result of subtracting the corresponding average value at each pixel position of the template; x, y represent the horizontal and vertical coordinates of the target image pixels; W, H represent the width and length of the image; x′, x″ represent the horizontal coordinates of the template pixels; y′, y″ represent the vertical coordinates of the template pixels; w, h represent the width and length of the template; where x∈[0, Ww], y∈[0, Hh], x′, x″∈[0, w-1], y′, y″∈[0, h-1].
9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the pure vision-based track extraction method according to any one of claims 1 to 4 are implemented.
10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the computer program is executed by a processor, the steps of the pure vision-based track extraction method according to any one of claims 1 to 4 are implemented.
Citation Information
Patent Citations
Lane line detection method based on vanishing point estimation and semantic segmentation
CN111582083A
Track line fitting method in complex scene based on image segmentation
CN112800902A
Railway route information acquisition device and railway route information acquisition method
CN115916623A
Visual perception method for three-dimensional space line shape of large-area railway track
CN117542038A
Cited By
Pure vision vehicle positioning method and system based on track identification
CN121330581A
