A visual calibration and parameter correction method and system
Through the feature point detection and matrix replacement of SuperPoint and LightGlue networks, combined with period matching and judgment strategies, the automatic calibration and high-precision correction of the camera are achieved, and the calibration failure problem caused by camera angle offset is solved, ensuring vehicle-road collaborative application High-precision visual perception capability.
Patent Information
- Application Number
- CN202410632935.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-21
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-05-21
AI Technical Summary
During use, the angle offset caused by factors such as climate, temperature changes, mechanical vibration and other factors affects the picture accuracy of the camera output and the accuracy of the visual processing algorithm, resulting in calibration failure or reduction in accuracy, and the inability to effectively monitor and dynamics. Corrected.
The SuperPoint network and LightGlue network are used to automatically calibrate and correct the camera. By obtaining the feature points of the initial calibration image and real-time detection of the image, calculating the homography matrix, replacing the perspective transformation matrix, realizing image calibration correction, and using periodic matching strategies and features The determination strategy of the number of points and distance difference, dynamically adjust the correction frequency.
It realizes automatic calibration and high-precision correction of the camera, reduces frequent calibration operations, ensures that the camera is always in a precise state, improves visual perception capabilities, and reduces operating costs and computing power consumption.
Smart Images

Figure CN118505816B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent driving and vehicle-road collaboration technology, and specifically to a visual calibration and parameter correction method and system. Background Art
[0002] In modern intelligent transportation systems, especially in the vehicle-road collaboration industry, cameras, as key perception devices, play a core role in perception and recognition. However, to ensure the effective implementation and accurate execution of these functions, cameras must be calibrated with highly accurate external parameters to ensure the accuracy of the conversion between their image coordinate system and geographic coordinate system.
[0003] The camera extrinsic calibration method is to obtain the perspective transformation matrix through the pixel coordinates of several points in the image and the longitude and latitude coordinates of these points. After obtaining the perspective transformation matrix, the pixel coordinates of the target on the image can be transformed through the matrix to obtain the position of the target in the real geographic coordinate system.
[0004] The technical problem with existing calibration methods is that, as existing cameras are used over time, especially in complex road environments, they are affected by factors such as climate, temperature changes, mechanical vibrations, and long-term stress release, which may cause a slight deviation in the initial installation angle of the camera. Such angle deviation will cause unexpected changes in the image output by the camera, which will seriously affect the accuracy of distance measurement, object positioning, and other visual processing algorithms based on camera data, and ultimately make the camera calibration that has been completed invalid or reduce its accuracy.
[0005] Therefore, in order to maintain high-precision visual perception capabilities in vehicle-road cooperative applications, it is urgent to develop an effective method that can monitor and dynamically correct the camera angle offset problems caused by the various reasons mentioned above, to ensure that the camera is always in a precisely calibrated state without the need for frequent recalibration operations. Summary of the invention
[0006] The present invention aims to provide a method and system for visual calibration and parameter correction, which can realize automatic calibration and correction of a camera with high correction accuracy.
[0007] To achieve the above object, the basic scheme provided by the present invention is:
[0008] Solution 1
[0009] A visual calibration and parameter correction method comprises the following steps:
[0010] Step 1: Get the initial calibration image A and real-time detection image B of the camera; and retain the homography matrix H corresponding to the initial calibration image A and the UTM coordinate system AC ;
[0011] Step 2, using the SuperPoint network to extract feature points and descriptors of the initial calibration image A and the real-time detection image B;
[0012] Step 3: Based on the feature points of the initial calibration image A and the real-time detection image B, the homography matrix H is obtained. AB ;
[0013] Step 4, combine H AC and H AB , solve and obtain the homography matrix H corresponding to the real-time detection image B and the UTM coordinate system BC ;
[0014] Step 5, use H BC Replace H AC , complete the image calibration and correction of the camera.
[0015] Furthermore, in step 5, after completing the calibration and correction of the camera image, the current detection image of the camera is also saved as a reference image.
[0016] Furthermore, it also includes:
[0017] Step 6: At the target time point, use the SuperPoint network to detect the current detection image and the reference image of the camera and extract the feature points of the two; determine whether the feature points of the current detection image and the reference image are the same in the pixel coordinate dimension, and count the number of feature points that are different between the current detection image and the reference image;
[0018] Step 7: If the number of different feature points is greater than the preset threshold, it is determined that the camera angle is offset, and steps 1 to 5 are repeated to perform image calibration correction; if the number of different feature points is less than the preset threshold, the process ends.
[0019] Furthermore, it also includes:
[0020] Step 6: At the target time point, the current detection image and the reference image of the SuperPoint network detection camera are used to extract the feature points of the same type of detection target in the two images; and the distance difference between the feature points of the current detection image and the reference image in the pixel coordinate dimension is calculated;
[0021] Step 7: If the distance difference is greater than the preset threshold, it is determined that the camera angle is offset, and steps 1 to 5 are repeated to perform image calibration correction; if the distance difference is less than the preset threshold, the process ends.
[0022] Furthermore, the target time point is set according to a cycle matching strategy, and the cycle matching strategy includes:
[0023] The positioning error value is set based on the minimum detection target of traffic participants in the camera's sensing range; after acquiring the reference image, the current detection image of the camera is collected again after an interval of 1 hour, and the feature points set in the image are mapped to the coordinates of the UTM coordinate system through the perspective transformation matrix when calculating the original calibration; and the positioning coordinates of the feature points corresponding to the current detection image in the UTM coordinate system, and the distance difference between the two is calculated;
[0024] If the distance difference is less than the positioning error value, the subsequent target time point is set to every 1 hour, that is, the verification operation of steps 6 to 7 is triggered once every 1 hour; if the distance difference is greater than the positioning error value, the subsequent target time point is set to every 0.5 hour, that is, the verification operation of steps 6 to 7 is triggered once every 0.5 hour.
[0025] Further, in solving the homography matrix H AB Before, the LightGlue network is used to match feature points based on the descriptors of the initial calibration image A and the real-time detection image B, so as to solve the homography matrix H AB .
[0026] Furthermore, when repeating steps 1 to 5, the current detection image is used as the real-time detection image B.
[0027] Furthermore, in a straight road scenario, the preset threshold is 0 to 1 m; in a curved road scenario, the preset threshold is 0 to 1.2 m.
[0028] Furthermore, the preset threshold is a dynamic threshold; the dynamic threshold is dynamically adjusted according to the density of feature points in the reference image.
[0029] Solution 2
[0030] A visual calibration and parameter correction system, used to execute a visual calibration and parameter correction method as described in Scheme 1; comprising an acquisition module, a data processing module and a calibration correction module;
[0031] The acquisition module is used to obtain the initial calibration image A and the real-time detection image B of the camera; the data processing module is used to solve the homography matrix H corresponding to the initial calibration image A and the UTM coordinate system AC , the homography matrix H of the initial calibration image A and the real-time detection image B AB , and the homography matrix H corresponding to the real-time detection image B and the UTM coordinate system BC ; The calibration correction module is used to use H BC Replace H AC , complete the image calibration and correction of the camera.
[0032] The working principle and advantages of the present invention are:
[0033] First, this solution can complete the image calibration and correction of the camera by automatically detecting and comparing the feature point transformations of the initial calibration image A and the real-time detection image B, and using the transformation and replacement of the homography matrix. The operation process is simple and can be automated, with high calibration and correction efficiency. In addition, this solution can be used to automatically perform daily inspections on the camera, accurately and promptly identify whether the camera has an offset problem and automatically complete the calibration process, ensuring that the camera is always in an accurately calibrated state without the need to frequently re-calibrate, with low computing power consumption and low operating costs.
[0034] Second, this solution uses the SuperPoint network and LightGlue network based on deep learning to detect and match image feature points, which can achieve a high correction accuracy. Compared with the feature point detection algorithms such as SIFT and SURF commonly used in existing image comparison solutions, the SuperPoint network selected in this solution has scale and rotation invariance, and can directly implement these characteristics in the neural network architecture, thus showing better generalization and robustness in some cases. The trained SuperPoint network can generate high-quality descriptors and increase the number of feature points, so that the matching accuracy can be improved, thereby improving the correction accuracy. The LightGlue network also has better matching accuracy than the KNN network. In addition, this solution provides judgment based on the number of feature points and judgment based on distance difference. These two different correction judgment strategies can meet different calibration correction needs. The former is more accurate and the latter is more comprehensive. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Figure 1 The present invention is a method flow chart of a visual calibration and parameter correction method and system embodiment 1. DETAILED DESCRIPTION
[0036] The following is a further detailed description through specific implementation methods:
[0037] Embodiment 1
[0038] The embodiment is basically as shown in the attached Figure 1 As shown: A visual calibration and parameter correction method, comprising the following steps:
[0039] Step 1: Get the initial calibration image A and real-time detection image B of the camera; and retain the homography matrix H corresponding to the initial calibration image A and the UTM coordinate system C AC The UTM coordinate system refers to the geographic coordinate system.
[0040] Where C = A·H AC; A represents the point set on plane A (initial calibration image), which is a set of pixel coordinates of a two-dimensional image; C represents the point set in the UTM coordinate system. AC is the homography matrix from plane A to UTM coordinate system C.
[0041] Furthermore, the camera in this embodiment is a perception camera used for vehicle-road collaboration.
[0042] Specifically, it can be a perception camera mounted on a roadside device. Furthermore, since the camera is installed at a fixed position at the end of the road, the initial calibration image is consistent with the background information in the real-time detection image, and the foreground information is often inconsistent because the target object in the visual perception area moves in real time. Here, when calculating the homography matrix, the calibration features are constructed based on the background information in the image, and the spatial mapping transformation matrix (i.e., homography matrix) from the image pixels to the UTM coordinate system is established, and each pixel corresponds to a UTM coordinate position.
[0043] Step 2: Use the SuperPoint network to extract feature points and descriptors of the initial calibration image A and the real-time detection image B.
[0044] Here, the descriptor generated by the SuperPoint network is a compact and discriminative numerical vector that is used to characterize the local image information around the feature point.
[0045] Each detected feature point is assigned a fixed length (usually 256 dimensions) floating point vector (i.e., descriptor). This descriptor can reflect the local features of the feature point in the image, such as texture, edge, and structure. In this SuperPoint network, descriptor generation is achieved through a convolutional neural network, which processes the image in the neighborhood of the feature point and outputs the descriptor.
[0046] This setting allows the corresponding feature points on the same object surface in different images to generate similar descriptors even when the lighting, viewing angle or image noise changes, so that they can be effectively matched and corresponded. The descriptor generation process and content are automatically learned and optimized by the network to ensure stability and robustness under a variety of complex transformations.
[0047] Furthermore, in this embodiment, a specific data set is used to pre-train the SuperPoint network to improve its extraction precision and accuracy, and ensure accurate and comprehensive detection of feature points and descriptors. The specific data set includes historical detection images of the camera.
[0048] Step 3: Based on the feature points of the initial calibration image A and the real-time detection image B, the homography matrix H is obtained. AB .
[0049] In this step, when solving the homography matrix H AB Before, the LightGlue network is used to match feature points based on the descriptors of the initial calibration image A and the real-time detection image B, so as to solve the homography matrix H AB .
[0050] Where B = A·H AB ; A and B represent the point sets on plane A (initial calibration image) and plane B (real-time detection image), respectively, and both are pixel coordinate sets of a two-dimensional image. AB is the homography matrix from plane A to plane B; it is a 3x3 matrix that describes how to project or map all points on plane A onto plane B.
[0051] Step 4, combine H AC and H AB , solve and obtain the homography matrix H corresponding to the real-time detection image B and the UTM coordinate system BC .
[0052] H BC =H AC ·(H AB ) -1 ; Among them, H BC is the homography matrix from plane B (real-time detection image) to UTM coordinate system C; (H AB ) -1 H AB The inverse matrix of .
[0053] Step 5, use H BC Replace H AC , complete the image calibration and correction of the camera.
[0054] In this step, after completing the calibration and correction of the camera image, the current detection image of the camera is also saved as a reference image.
[0055] Step 6: At the target time point, use the SuperPoint network to detect the camera's current detection image and reference image and extract feature points of the same type of detection target in the two; and calculate the distance difference between the feature points of the current detection image and the reference image in the pixel coordinate dimension.
[0056] The target time point can be selected as the same time point every day. In this embodiment, a time point is randomly selected in the afternoon as a fixed time point, and the camera is calibrated and corrected every day.
[0057] Preferably, the target time point can also be set according to a period matching strategy. The period matching strategy includes: taking the minimum detection target of traffic participants in the camera sensing range as a benchmark, setting a positioning error value, where the minimum detection target is selected as a pedestrian, and the corresponding positioning error value is set to 0.5m.
[0058] After acquiring the reference image, the current detection image of the camera is collected again after an interval of 1 hour. When calculating the original calibration (i.e., the calibration stage of steps 1 to 5), the feature points set in the image (here, the feature points of the minimum detection target are selected) are mapped to the coordinates of the UTM coordinate system through the perspective transformation matrix; and the current detection image corresponds to the positioning coordinates of the feature points in the UTM coordinate system, and the distance difference between the two is calculated. If the distance difference is less than the positioning error value (i.e., within the range of 0 to 0.5m), the subsequent target time points are set to every 1 hour, that is, the verification operation of steps 6 to 7 is triggered once every 1 hour. If the distance difference is greater than the positioning error value (i.e., more than 0.5m), the subsequent target time points are set to every 0.5 hour, that is, the verification operation of steps 6 to 7 is triggered once every 0.5 hour.
[0059] Compared with the calibration at a fixed time point, this setting is more proactive and can detect the camera offset in time, trigger calibration and correction in time, and reduce the impact of camera offset on the perceived data to a greater extent. In addition, compared with the real-time verification method, this method can avoid the computing power consumption caused by real-time matching correction calculations while ensuring timely correction, and has lower operating costs. In addition, by deploying this solution in batches, automatic calibration and correction can be achieved for all intersection scenarios, reducing the cost of road-side calibration and correction in the field of intelligent driving and vehicle-road collaboration.
[0060] Step 7: If the distance difference is greater than the preset threshold, it is determined that the camera angle is offset, and steps 1 to 5 are repeated to perform image calibration correction; if the distance difference is less than the preset threshold, the process ends.
[0061] When steps 1 to 5 are repeated, the current detection image is used as the real-time detection image B.
[0062] Specifically, in practical applications, the preset threshold is set to different numerical ranges based on factors such as the actual structured road size, road slope (due to different mapping relationships between uphill and downhill slopes, the corresponding real-world distances between adjacent pixels are different), and image pixels (different coordinate systems, different distance lengths between adjacent pixel coordinates at long distances and short distances). In addition, since the turning section of the structured road will span multiple vertical and horizontal pixel areas, and the density changes of the pixel areas spanned by the straight section are inconsistent, during the original calibration (i.e., when calibrating based on the initial calibration image A and the real-time detection image B), a local calibration of the vehicle in the road turning area will be established, that is, a preset threshold will be set separately for the local adjacent pixel areas in the road turning area.
[0063] In this embodiment, in a straight road scenario (that is, when the scene within the perception range of the camera belongs to a straight road scenario), the preset threshold is 0 to 1m; in a curved road scenario (that is, when the scene within the perception range of the camera belongs to a curved road scenario), the preset threshold is 0 to 1.2m. Specifically, in a straight road scenario, for the long-distance scene therein, the preset threshold is further set to 0.5 to 1m; for the close-distance scene therein, the preset threshold is further set to 0 to 0.5m. In a curved road scenario, for the long-distance scene therein, the preset threshold is further set to 0.6 to 1.2m; for the long-distance scene therein, the preset threshold is further set to 0 to 0.6m. Optionally, the criterion for distinguishing between short distance and long distance may be that, based on the installation position of the camera, the first 50% area of the camera's perception range close to the camera's installation position is the short distance, and the last 50% area is the long distance. With this setting, the solution can effectively identify the camera's offset and dynamically correct it in time, ensuring that the camera is always in an accurately calibrated state without repeatedly adjusting the camera's hardware structure. In addition, the solution sets different preset thresholds for the perception area scenes that different cameras are responsible for, which can ensure accurate offset determination.
[0064] This embodiment also provides a visual calibration and parameter correction system, which is used to execute a visual calibration and parameter correction method as described in Scheme 1; it includes an acquisition module, a data processing module and a calibration correction module;
[0065] The acquisition module is used to obtain the initial calibration image A and the real-time detection image B of the camera; the data processing module is used to solve the homography matrix H corresponding to the initial calibration image A and the UTM coordinate system AC , the homography matrix H of the initial calibration image A and the real-time detection image B AB , and the homography matrix H corresponding to the real-time detection image B and the UTM coordinate system BC ; The calibration correction module is used to use H BC Replace HAC , complete the image calibration and correction of the camera.
[0066] The present embodiment provides a method and system for visual calibration and parameter correction, which can realize automatic calibration and correction of a camera with high correction accuracy.
[0067] Embodiment 2
[0068] A visual calibration and parameter correction method is different from the first embodiment in that:
[0069] Step 6: At the target time point, use the SuperPoint network to detect the current detection image and the reference image of the camera and extract the feature points of the two; and determine whether the feature points of the current detection image and the reference image are the same in the pixel coordinate dimension, and count the number of feature points that are different between the current detection image and the reference image.
[0070] The target time point may be selected as the same time point every day, or may be set according to a periodic matching strategy.
[0071] Step 7: If the number of different feature points is greater than the preset threshold, it is determined that the camera angle is offset, and steps 1 to 5 are repeated to perform image calibration correction; if the number of different feature points is less than the preset threshold, the process ends.
[0072] In this embodiment, the preset threshold is set to 1 / 4 of the total number of feature points in the reference image.
[0073] The present embodiment provides a method and system for visual calibration and parameter correction, which adopts a correction judgment strategy based on the number of feature points. Compared with the first embodiment, the present method is more suitable for coarse calibration scenarios and can efficiently complete correction judgment.
[0074] Embodiment 3
[0075] A visual calibration and parameter correction method, based on the first or second embodiment, in step 7, the preset threshold is a dynamic threshold; the dynamic threshold is dynamically adjusted according to the density of feature points in the reference image.
[0076] Specifically, the dynamic threshold is proportional to the density of feature points in the reference image. The greater the density, the greater the dynamic threshold.
[0077] In addition, dynamic thresholds are set for straight-ahead scenes and curved scenes respectively; in straight-ahead scenes, the reference value of the dynamic threshold is set to 0-1m; in curved scenes, the reference value of the dynamic threshold is set to 0-1.2m. In addition, in straight-ahead scenes, the reference value is further set to 0.5-1m for long-distance scenes; and the reference value is further set to 0-0.5m for close-distance scenes. In curved scenes, the reference value is further set to 0.6-1.2m for long-distance scenes; and the reference value is further set to 0-0.6m for long-distance scenes.
[0078] By adopting this method, different thresholds are set for different pixel areas and different feature point densities to address the inconsistent changes in pixel density of targets in different scenes and the corresponding inconsistency in feature point extraction density. By setting dynamic thresholds for vehicle turning scenes and vehicle straight driving scenes respectively, the stability of feature point recognition can be further guaranteed.
[0079] The present embodiment provides a method and system for visual calibration and parameter correction, which can adapt to different camera image acquisition accuracies, and has a high degree of feature point comparison, thereby achieving a higher correction accuracy.
[0080] Embodiment 4
[0081] A visual calibration and parameter correction method is provided, based on the first or second embodiment, in which the setting method of the target time point is adjusted. The target time point can also be set according to the offset prediction strategy.
[0082] The offset prediction strategy includes: collecting basic parameters of the camera structure itself, and monitoring whether target environmental information appears at the camera installation location; the target environmental information includes weather information such as snowfall and strong wind that are likely to affect the physical installation angle of the camera, and event information such as a vehicle (or other object) colliding with the pole where the camera is located (or other supporting body where the camera is located) that is likely to affect the physical installation angle of the camera. In practical applications, the event information can be identified and obtained through cameras on surrounding poles to correct the camera. The basic parameters include the installation time of the camera.
[0083] When the target environment information appears, the time point is taken as a target time point, and steps 6 to 7 are performed to realize active calibration correction of the camera. In addition, when the camera has been installed for more than a certain number of years (such as 1 year), the time point is also taken as a target time point, and steps 6 to 7 are performed to realize active calibration correction of the camera.
[0084] A visual calibration and parameter correction method and system provided in this embodiment can capture possible installation angle deviation of the camera in combination with target environment information, and can correct the camera in time when the deviation occurs, with high timeliness and low computing power consumption.
[0085] Embodiment 5
[0086] A visual calibration and parameter correction method, based on the first or second embodiment, after completing an image calibration correction in step 7, can also include step 8: collecting the current detection image of the camera again, and selecting a fixed background information in the current detection image as a feature point, calculating the positioning coordinates of the feature point in the UTM coordinate system; and receiving the vehicle's recognition result of the fixed background information from the vehicle's onboard OBU passing through the camera sensing area and converting it into the positioning coordinates in the UTM coordinate system. And calculate the distance difference between the two positioning coordinates, if the distance difference is less than the positioning error value (here it can be set to 0.5m), it is determined that the calibration is accurate, and the process ends; if the distance difference is greater than the positioning error value, it is determined that the calibration is wrong, and repeat steps 1 to 5 again to perform image calibration correction.
[0087] A visual calibration and parameter correction method and system provided in this embodiment can perform a secondary verification of the calibration correction effect in combination with third-party (vehicle-side) data. The verification is more accurate and can further ensure that the camera correction is in place, which helps to improve the accuracy of subsequent perception data used for vehicle-road collaboration.
[0088] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is not described in detail here. The ordinary technicians in the relevant field are aware of all the common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement the scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for the ordinary technicians in the relevant field to implement this application. It should be pointed out that for the technicians in this field, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent.
Claims
1. A visual calibration and parameter correction method, characterized in that: The following steps are involved: Step 1: Get the camera's initial calibration image A and real-time detection image B; and retain the homography matrix corresponding to the initial calibration image A and the UTM coordinate system. ; Step 2, using the SuperPoint network to extract feature points and descriptors of the initial calibration image A and the real-time detection image B; Step 3: Based on the feature points of the initial calibration image A and the real-time detection image B, the homography matrix is obtained. ; In solving the homography matrix Before the image is extracted, the LightGlue network is used to match feature points based on the descriptors of the initial calibration image A and the real-time detection image B to solve the homography matrix. ; Step 4, Combine and , solve and obtain the homography matrix corresponding to the real-time detection image B and the UTM coordinate system ; Step 5, use replace , complete the image calibration and correction of the camera.
2. A visual calibration and parameter correction method according to claim 1, characterized in that: In step 5, after completing the calibration and correction of the camera image, the current detection image of the camera is also saved as a reference image.
3. A visual calibration and parameter correction method according to claim 2, characterized in that: Also includes: Step 6: At the target time point, the SuperPoint network is used to detect the current detection image and the reference image of the camera and extract feature points of the two images; And determine whether the feature points of the current detection image and the reference image are the same in the pixel coordinate dimension, and count the number of feature points that are different between the current detection image and the reference image; Step 7: If the number of different feature points is greater than the preset threshold, it is determined that the camera angle is offset, and steps 1 to 5 are repeated to perform image calibration correction; if the number of different feature points is less than the preset threshold, the process ends.
4. A visual calibration and parameter correction method according to claim 2, characterized in that: Also includes: Step 6: At the target time point, the current detection image and the reference image of the SuperPoint network detection camera are detected and feature points of the same type of detection target are extracted from the two images; And calculate the distance difference between the feature points of the current detection image and the reference image in the pixel coordinate dimension; Step 7: If the distance difference is greater than a preset threshold, it is determined that the camera angle is offset, and steps 1 to 5 are repeated to perform image calibration correction; If the distance difference is less than the preset threshold, the process ends.
5. A visual calibration and parameter correction method according to claim 3 or claim 4, characterized in that: The target time point is set according to a cycle matching strategy, and the cycle matching strategy includes: The positioning error value is set based on the minimum detection target of traffic participants in the camera's sensing range; after acquiring the reference image, the current detection image of the camera is collected again after an interval of 1 hour, and the feature points set in the image are mapped to the coordinates of the UTM coordinate system through the perspective transformation matrix when calculating the original calibration; and the positioning coordinates of the feature points corresponding to the current detection image in the UTM coordinate system, and the distance difference between the two is calculated; If the distance difference is less than the positioning error value, the subsequent target time point is set to every 1 hour, that is, the verification operation of steps 6 to 7 is triggered once every 1 hour; if the distance difference is greater than the positioning error value, the subsequent target time point is set to every 0.5 hour, that is, the verification operation of steps 6 to 7 is triggered once every 0.5 hour.
6. A visual calibration and parameter correction method according to claim 3 or claim 4, characterized in that: When steps 1 to 5 are repeated, the current detection image is used as the real-time detection image B.
7. A visual calibration and parameter correction method according to claim 4, characterized in that: In a straight road scenario, the preset threshold is 0~1m; in a curved road scenario, the preset threshold is 0~1.2m.
8. A visual calibration and parameter correction method according to claim 3 or claim 4, characterized in that: The preset threshold is a dynamic threshold; the dynamic threshold is dynamically adjusted according to the density of feature points in the reference image.
9. A visual calibration and parameter correction system, characterized in that: Used to execute a visual calibration and parameter correction method as described in any one of claims 1 to 8; comprising an acquisition module, a data processing module and a calibration correction module; The acquisition module is used to obtain the initial calibration image A and the real-time detection image B of the camera; the data processing module is used to solve the homography matrix corresponding to the initial calibration image A and the UTM coordinate system , the homography matrix of the initial calibration image A and the real-time detection image B , and the homography matrix corresponding to the real-time detection image B and the UTM coordinate system ; The calibration correction module is used to use replace , complete the image calibration and correction of the camera.
