Vehicle-mounted camera posture calculation method, device and computer-readable storage medium
By processing image frames through the feature extraction network and the NetVLAD network, and screening reference images in the scene database, combined with the local odometry and P3P pose estimation algorithm, the problem of low pose calculation accuracy of the on-board camera is solved, and more accurate obstacle positioning is achieved.
Patent Information
- Application Number
- CN202310904859.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-07-21
- Publication Date
- 2025-09-30
- Estimated Expiration
- 2043-07-21
AI Technical Summary
Existing methods for calculating the pose of vehicle-mounted cameras suffer from poor calculation accuracy and sparse feature point detection, making it difficult to effectively locate obstacles.
The feature extraction network model and the NetVLAD network model are used to process the image frames collected by the vehicle-mounted camera. By screening reference scene images from the pre-built scene database, combining the local odometry mapping principle and the P3P pose estimation algorithm, the rotation and translation matrices of the vehicle-mounted camera are calculated.
The accuracy of vehicle-mounted camera pose calculation and the generalization ability of dense feature maps are improved, achieving more accurate obstacle positioning.
Smart Images
Figure CN116883966B_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to the technical field of vehicle-mounted camera image processing, and in particular to a vehicle-mounted camera posture calculation method, device, and computer-readable storage medium. Background Art
[0002] At present, visual positioning of vehicle-mounted cameras based on image feature matching (also known as: pose estimation or posture calculation) is a relatively common processing method. Among them, image feature matching is a more critical step, which can effectively solve the data association problem in visual positioning. Therefore, the accuracy of image feature matching largely determines the effect of visual positioning.
[0003] Existing image feature matching methods mainly use the SIFT (Scale-Invariant Feature Transform) algorithm or the ORB (Oriented Fast and Rotated Brief) algorithm, and the main steps include feature extraction, feature description, and feature matching. However, during the specific implementation, the inventors found that the above-mentioned image feature matching methods have many defects. For example, the reconstruction results obtained by the feature point-based method are relatively sparse, and it is difficult to detect a sufficient number of feature points when applied in real environments. Secondly, when applied to the posture calculation of the vehicle-mounted camera, the posture calculation accuracy is relatively poor, which is not conducive to the positioning of obstacles by image vision. Summary of the Invention
[0004] The technical problem to be solved by the embodiments of the present invention is to provide a method for calculating the posture of a vehicle-mounted camera, which can effectively improve the calculation accuracy.
[0005] A further technical problem to be solved by the embodiments of the present invention is to provide a vehicle-mounted camera posture calculation device that can effectively improve calculation accuracy.
[0006] A further technical problem to be solved by an embodiment of the present invention is to provide a computer-readable storage medium for storing a computer program that can effectively improve the accuracy of posture calculation of a vehicle-mounted camera.
[0007] In order to solve the above technical problems, the embodiment of the present invention provides the following technical solution: a method for calculating the posture of a vehicle-mounted camera, comprising the following steps:
[0008] Extracting a current image frame from an original image of the vehicle's surroundings captured and transmitted in real time by an on-board camera, and sequentially processing the current image frame through a pre-stored feature extraction network model and a NetVLAD network model to obtain a global descriptor of the current image frame, wherein the feature extraction network model is composed of a plurality of feature extraction layers sequentially connected;
[0009] Filtering a reference scene image that matches the global descriptor from a pre-stored scene database, wherein the scene database is pre-built scene map data of a motor vehicle in an actual driving environment;
[0010] Upsampling the current image frame using each feature extraction layer of the feature extraction network model to obtain corresponding features to be fused, adding and fusing each of the features to be fused and then processing them using a nonlinear activation function to obtain an actual dense high-dimensional feature map of the current image frame;
[0011] Determine a region of interest in the reference scene image, calculate matching feature points in the actual dense high-dimensional feature map that correspond one-to-one to each key feature point in the region of interest, wherein an actual dot product of the same key feature point and the corresponding matching feature point is greater than an actual dot product of the same key feature point and the corresponding matching feature point with respect to the remaining feature points in the actual dense high-dimensional feature map, and an actual distance between the same key feature point and the corresponding matching feature point is less than an actual distance between the same key feature point and the corresponding matching feature point with respect to the remaining feature points in the actual dense high-dimensional feature map; and
[0012] The current rotation and translation matrix of the vehicle-mounted camera is calculated using the mutually matching key feature points and matching feature points based on the local odometer mapping principle and the P3P pose estimation algorithm.
[0013] Furthermore, the filtering out of a reference scene image matching the global descriptor from a pre-stored scene database specifically includes:
[0014] Calculating the actual Hamming distance between the global descriptor and the VLAD vector of each key frame in the scene database, and selecting key frames from the scene database that meet a predetermined preliminary screening condition based on a decision tree algorithm model, wherein the predetermined preliminary screening condition is that the descending sorting sequence number of the actual Hamming distance is less than or equal to a predetermined sequence number; and
[0015] The actual similarity between the adjacent image frames of each key frame that meets the predetermined preliminary screening condition and the current image frame is calculated, and the key frame having the maximum actual similarity and meeting the predetermined preliminary screening condition is determined as the reference scene image.
[0016] Furthermore, the global descriptor is first subjected to dimensionality reduction processing based on a principal component analysis algorithm, and then the actual Hamming distance between the global descriptor after dimensionality reduction processing and the VLAD vector of each key frame in the scene database is calculated.
[0017] Furthermore, the method of using the mutually matched key feature points and matching feature points to calculate the current rotation and translation matrix of the vehicle-mounted camera based on the local odometer mapping principle and the P3P pose estimation algorithm specifically includes:
[0018] Construct the actual three-dimensional space of the vehicle driving scene based on the principle of local odometry mapping;
[0019] Projecting the key feature points and the matching feature points that are matched to each other into the actual three-dimensional space to generate key three-dimensional points and matching three-dimensional points respectively; and
[0020] The current rotation and translation matrix of the vehicle-mounted camera is calculated based on the P3P pose estimation algorithm using three pairs of non-collinear and mutually matching key three-dimensional points and matching three-dimensional points.
[0021] Furthermore, the method further comprises:
[0022] Calculating actual projection errors of the remaining mutually matching key three-dimensional points and matching three-dimensional points corresponding to the rotation and translation matrices based on a random sampling algorithm model; and
[0023] The minimum value of the actual projection error is calculated based on the error optimization algorithm model, and the rotation and translation matrix is recalculated and updated according to the key three-dimensional points and matching three-dimensional points that meet the minimum value and match each other.
[0024] Furthermore, the region of interest is determined by judging whether the actual number of corner points in each region of the reference scene image is greater than a predetermined number.
[0025] On the other hand, in order to solve the above-mentioned technical problems, an embodiment of the present invention provides the following technical solutions: a vehicle-mounted camera posture calculation device, connected to a vehicle-mounted camera of a motor vehicle, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the vehicle-mounted camera posture calculation method as described in any one of the above-mentioned items.
[0026] On the other hand, in order to solve the above-mentioned technical problems, an embodiment of the present invention provides the following technical solution: a computer-readable storage medium, comprising a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the vehicle-mounted camera posture calculation method as described in any one of the above-mentioned items.
[0027] After adopting the above technical solution, the embodiment of the present invention has at least the following beneficial effects: the embodiment of the present invention first obtains a global descriptor by processing the current image frame acquired by the vehicle-mounted camera through the feature extraction network model and the NetVLAD network model, and further determines the reference scene image matching the global descriptor by screening in the scene database. Due to the pre-constructed scene map data of the motor vehicle in the actual driving environment, it is ensured that the global descriptor matches the actual driving environment of the motor vehicle as much as possible; further, each feature extraction layer of the feature extraction network model is used to upsample the current image frame to obtain the corresponding features to be fused, and each of the features to be fused is added and fused, and then processed using a nonlinear activation function to obtain an actual dense high-dimensional feature map. The actual dense high-dimensional feature map not only fuses Features of different scales are processed through nonlinear activation functions to improve the nonlinear description of key points by high-dimensional features, so that the generalization ability of actual dense high-dimensional feature maps is improved; further, by calculating the matching feature points in the actual dense high-dimensional feature map that correspond one-to-one to each key feature point in the region of interest, and using dot product matching and actual distance to perform screening and judgment, the mutual correlation between the key feature points and each feature point in the actual dense high-dimensional feature map is determined, and the matching feature points that match the key feature points are screened out based on the maximum correlation principle; finally, the current rotation and translation matrix of the vehicle-mounted camera can be calculated based on the local odometer mapping principle and the P3P pose estimation algorithm using the mutually matching key feature points and matching feature points, so as to realize the posture calculation of the vehicle-mounted camera with higher calculation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 The figure is a flowchart of an optional embodiment of the method for calculating the posture of a vehicle-mounted camera of the present invention.
[0029] Figure 2 This is a specific flow chart of step S2 of an optional embodiment of the method for calculating the posture of a vehicle-mounted camera of the present invention.
[0030] Figure 3 This is a specific flow chart of step S5 of an optional embodiment of the method for calculating the posture of a vehicle-mounted camera of the present invention.
[0031] Figure 4 This is a functional block diagram of an optional embodiment of the vehicle-mounted camera posture calculation device of the present invention.
[0032] Figure 5 This is a functional module diagram of an optional embodiment of the vehicle-mounted camera posture calculation device of the present invention. DETAILED DESCRIPTION
[0033] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the following exemplary embodiments and descriptions are only used to explain the present invention and are not intended to limit the present invention. Moreover, the embodiments and features in the embodiments of the present application may be combined with each other unless there is a conflict.
[0034] like Figure 1 As shown, an optional embodiment of the present invention provides a method for calculating the posture of a vehicle-mounted camera, comprising the following steps:
[0035] S1: extracting a current image frame from an original image of the vehicle's surroundings captured and transmitted in real time by the vehicle-mounted camera 1, and processing the current image frame sequentially through a pre-stored feature extraction network model and a NetVLAD network model to obtain a global descriptor of the current image frame, wherein the feature extraction network model is composed of a plurality of feature extraction layers connected in sequence;
[0036] S2: Filtering a reference scene image that matches the global descriptor from a pre-stored scene database, wherein the scene database is pre-built scene map data of a motor vehicle in an actual driving environment;
[0037] S3: upsampling the current image frame using each feature extraction layer of the feature extraction network model to obtain corresponding features to be fused, adding and fusing each of the features to be fused and then processing them using a nonlinear activation function to obtain an actual dense high-dimensional feature map of the current image frame;
[0038] S4: determining a region of interest in the reference scene image, calculating matching feature points in the actual dense high-dimensional feature map that correspond one-to-one to each key feature point in the region of interest, wherein an actual dot product between the same key feature point and the corresponding matching feature point is greater than an actual dot product between the same key feature point and the corresponding matching feature point and the actual distance between the same key feature point and the corresponding matching feature point is less than an actual distance between the same key feature point and the corresponding matching feature point and the ...
[0039] S5: Using the mutually matching key feature points and matching feature points, the current rotation and translation matrix of the vehicle-mounted camera 1 is calculated based on the local odometer mapping principle and the P3P pose estimation algorithm.
[0040] In the embodiment of the present invention, the current image frame acquired by the vehicle-mounted camera 1 is first processed by the feature extraction network model and the NetVLAD network model to obtain a global descriptor, and further a reference scene image matching the global descriptor is determined by screening in the scene database. Due to the pre-constructed scene map data of the motor vehicle in the actual driving environment, the global descriptor is ensured to match the actual driving environment of the motor vehicle as much as possible; further, each feature extraction layer of the feature extraction network model is used to upsample the current image frame to obtain the corresponding features to be fused, and each of the features to be fused is added and fused, and then processed using a nonlinear activation function to obtain an actual dense high-dimensional feature map. The actual dense high-dimensional feature map not only fuses features of different scales, but also through nonlinear The activation function is used to improve the nonlinear description of key points by high-dimensional features, thereby improving the generalization ability of the actual dense high-dimensional feature map; further, by calculating the matching feature points in the actual dense high-dimensional feature map that correspond one-to-one to each key feature point in the area of interest, and using dot product matching and actual distance to perform screening and judgment, the mutual correlation between the key feature points and each feature point in the actual dense high-dimensional feature map is determined, and the matching feature points that match the key feature points are screened out based on the maximum correlation principle; finally, the current rotation and translation matrix of the vehicle-mounted camera 1 can be calculated based on the local odometer mapping principle and the P3P pose estimation algorithm using the mutually matching key feature points and matching feature points, so as to realize the posture calculation of the vehicle-mounted camera 1 with higher calculation accuracy.
[0041] In a specific implementation, the actual dense high-dimensional feature map is set as DHC_q, and the region of interest in the reference scene image is set as SHC_r, then the dot product of the two can be expressed as:
[0042] HC cross =SHC_r·DHC_q (Formula 1)
[0043] Among them, HC cross The dot product of the high-dimensional features representing the region of interest of the reference scene image and the actual dense high-dimensional feature map.
[0044] In an optional embodiment of the present invention, Figure 2 As shown, the step S2 specifically includes:
[0045] S21: calculating the actual Hamming distance between the global descriptor and the VLAD vector of each key frame in the scene database, and selecting key frames from the scene database that meet a predetermined preliminary screening condition based on a decision tree algorithm model, wherein the predetermined preliminary screening condition is that the descending sorting number of the actual Hamming distance is less than or equal to a predetermined number; and
[0046] S22: Calculate and obtain actual similarities between adjacent image frames of each key frame that meets the predetermined preliminary screening condition and the current image frame, and determine the key frame with the largest actual similarity and that meets the predetermined preliminary screening condition as the reference scene image.
[0047] In this embodiment, the actual Hamming distance between the global descriptor and the VLAD vector of each key frame in the scene database is first calculated. The larger the actual Hamming distance, the higher the similarity between the global descriptor and the key frame. Therefore, the key frames are sorted in descending order using the actual Hamming distance corresponding to each key frame, thereby screening out the key frames with the highest actual Hamming distance. Based on the principle that the similarity between the adjacent frames (i.e., the previous frame and the next frame) of the real reference scene image and the current image frame should be higher than the similarity between the adjacent frames of the fake reference scene image and the current image frame, the actual similarity between the adjacent image frames of the key frames that meet the predetermined preliminary screening conditions and the current image frame is determined, thereby further screening is performed to accurately determine the real reference scene image.
[0048] In specific implementation, it is understandable that by sorting the key frames in descending order with reference to the corresponding actual Hamming distances, the predetermined sequence number can be set to 10, that is, the key frames with actual Hamming distance sequence numbers of 1 to 10 are filtered out.
[0049] In an optional embodiment of the present invention, the global descriptor is first subjected to dimensionality reduction processing based on a principal component analysis (PCA) algorithm, and then the actual Hamming distance between the global descriptor after dimensionality reduction processing and the VLAD vectors of each key frame in the scene database is calculated. In this embodiment, by further performing dimensionality reduction processing on the global descriptor based on the PCA algorithm, the computational complexity of the global descriptor can be effectively reduced, thereby improving computational efficiency.
[0050] In specific implementation, the actual Hamming distance calculation formula is as follows:
[0051]
[0052] Among them, D hanming represents the actual Hamming distance, V Pca Represents the k-dimensional binary vector of the global descriptor after dimensionality reduction, M i Represents the k-dimensional binary global descriptor corresponding to the i-th key frame in the scene database.
[0053] In an optional embodiment of the present invention, Figure 3 As shown, the step S5 specifically includes:
[0054] S51: Constructing the actual 3D space of the vehicle driving scene based on the principle of local odometry mapping;
[0055] S52: Projecting the key feature points and the matching feature points that are matched to each other into the actual three-dimensional space to generate key three-dimensional points and matching three-dimensional points respectively; and
[0056] S53: three pairs of non-collinear and mutually matching key three-dimensional points and matching three-dimensional points are used to calculate and obtain the current rotation and translation matrix of the vehicle-mounted camera 1 based on the P3P pose estimation algorithm.
[0057] In this embodiment, the actual three-dimensional space of the motor vehicle driving scene is first constructed based on the local odometer mapping principle, and then the matched key feature points and the matching feature points are projected into the actual three-dimensional space in sequence. Finally, based on the P3P pose estimation algorithm and the key three-dimensional points and matching three-dimensional points generated by the projection, the current rotation and translation matrix of the vehicle-mounted camera 1 can be quickly calculated to realize the posture calculation. The calculation process is relatively simple.
[0058] In an optional embodiment of the present invention, the method further includes:
[0059] Calculating actual projection errors of the remaining mutually matching key three-dimensional points and matching three-dimensional points corresponding to the rotation and translation matrices based on a random sampling algorithm model; and
[0060] The minimum value of the actual projection error is calculated based on the error optimization algorithm model, and the rotation and translation matrix is recalculated and updated according to the key three-dimensional points and matching three-dimensional points that meet the minimum value and match each other.
[0061] In this embodiment, a random sampling algorithm model is also used to calculate the actual projection errors of the rotation and translation matrix corresponding to the remaining mutually matching key three-dimensional points and matching three-dimensional points, so that the minimum value of the actual projection error is calculated through the error optimization algorithm model. The rotation and translation matrix is recalculated and updated based on the key three-dimensional points and matching three-dimensional points that meet the minimum value and match each other, thereby optimizing the rotation and translation matrix and improving the calculation accuracy.
[0062] In an optional embodiment of the present invention, the region of interest is determined by determining whether the actual number of corner points in each region of the reference scene image is greater than a predetermined number. Generally, a region with more corner points has more features, which facilitates image matching. Therefore, in this embodiment, the region of interest is determined by determining the number of corner points in each region of the reference scene image. The determination principle is simple.
[0063] On the other hand, Figure 4As shown, an embodiment of the present invention provides a vehicle-mounted camera posture calculation device 3, which is connected to a vehicle-mounted camera 1 of a motor vehicle, and includes a processor 30, a memory 32, and a computer program stored in the memory and configured to be executed by the processor 30. When the processor 30 executes the computer program, the vehicle-mounted camera posture calculation method described in the above embodiment is implemented.
[0064] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory 32 and executed by the processor 30 to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program in the vehicle-mounted camera posture calculation device 3. For example, the computer program may be divided into Figure 5 The functional modules in the vehicle-mounted camera posture calculation device 3, wherein the image acquisition and descriptor extraction module 41, the scene graph screening module 42, the dense feature calculation module 43, the feature point matching module 44 and the posture calculation module 45 respectively execute the above steps S1 to S5.
[0065] The vehicle-mounted camera posture calculation device 3 can be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The vehicle-mounted camera posture calculation device 3 may include, but is not limited to, a processor 30 and a memory 32. Those skilled in the art will appreciate that the schematic diagram is merely an example of the vehicle-mounted camera posture calculation device 3 and does not limit the vehicle-mounted camera posture calculation device 3. The vehicle-mounted camera posture calculation device 3 may include more or fewer components than shown, or may combine certain components, or different components. For example, the vehicle-mounted camera posture calculation device 3 may also include input and output devices, network access devices, buses, etc.
[0066] The processor 30 may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor. The processor 30 is the control center of the vehicle-mounted camera posture calculation device 3 and connects various parts of the vehicle-mounted camera posture calculation device 3 using various interfaces and lines.
[0067] The memory 32 can be used to store the computer programs and / or modules. The processor 30 implements the various functions of the vehicle-mounted camera posture calculation device 3 by running or executing the computer programs and / or modules stored in the memory 32 and calling the data stored in the memory 32. The memory 32 may mainly include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a pattern recognition function and a pattern overlay function); the data storage area may store data (such as graphic data) generated based on the use of the vehicle-mounted camera posture calculation device 3. In addition, the memory 32 may include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.
[0068] If the functions described in the embodiments of the present invention are implemented in the form of software function modules or units and sold or used as independent products, they can be stored in a storage medium readable by a computing device. Based on this understanding, the embodiments of the present invention implement all or part of the processes in the above-mentioned embodiments, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor 30, it can implement the steps of the above-mentioned various method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.
[0069] On the other hand, an embodiment of the present invention provides a computer-readable storage medium, including a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the vehicle-mounted camera posture calculation method as described in the above embodiment.
[0070] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.
[0071] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which fall within the scope of protection of the present invention.
Claims
1. A method for calculating the posture of a vehicle-mounted camera, characterized in that: The method comprises the following steps: Extracting a current image frame from an original image of the vehicle's surroundings captured and transmitted in real time by an on-board camera, and sequentially processing the current image frame through a pre-stored feature extraction network model and a NetVLAD network model to obtain a global descriptor of the current image frame, wherein the feature extraction network model is composed of a plurality of feature extraction layers sequentially connected; Filtering a reference scene image that matches the global descriptor from a pre-stored scene database, wherein the scene database is pre-built scene map data of a motor vehicle in an actual driving environment; Upsampling the current image frame using each feature extraction layer of the feature extraction network model to obtain corresponding features to be fused, adding and fusing each of the features to be fused and then processing them using a nonlinear activation function to obtain an actual dense high-dimensional feature map of the current image frame; Determine a region of interest in the reference scene image, calculate matching feature points in the actual dense high-dimensional feature map that correspond one-to-one to each key feature point in the region of interest, wherein an actual dot product of the same key feature point and the corresponding matching feature point is greater than an actual dot product of the same key feature point and the corresponding matching feature point with respect to the remaining feature points in the actual dense high-dimensional feature map, and an actual distance between the same key feature point and the corresponding matching feature point is less than an actual distance between the same key feature point and the corresponding matching feature point with respect to the remaining feature points in the actual dense high-dimensional feature map; and The current rotation and translation matrix of the vehicle-mounted camera is calculated using the mutually matching key feature points and matching feature points based on the local odometer mapping principle and the P3P pose estimation algorithm.
2. The method for calculating the posture of a vehicle-mounted camera according to claim 1, wherein: The step of selecting a reference scene image that matches the global descriptor from a pre-stored scene database specifically includes: Calculating the actual Hamming distance between the global descriptor and the VLAD vector of each key frame in the scene database, and selecting key frames from the scene database that meet a predetermined preliminary screening condition based on a decision tree algorithm model, wherein the predetermined preliminary screening condition is that the descending sorting sequence number of the actual Hamming distance is less than or equal to a predetermined sequence number; and The actual similarity between the adjacent image frames of each key frame that meets the predetermined preliminary screening condition and the current image frame is calculated, and the key frame having the maximum actual similarity and meeting the predetermined preliminary screening condition is determined as the reference scene image.
3. The method for calculating the posture of a vehicle-mounted camera according to claim 2, wherein: First, the global descriptor is subjected to dimensionality reduction processing based on a principal component analysis algorithm, and then the actual Hamming distance between the global descriptor after dimensionality reduction processing and the VLAD vector of each key frame in the scene database is calculated.
4. The method for calculating the posture of a vehicle-mounted camera according to claim 1, wherein: The method of using the mutually matched key feature points and matching feature points to calculate the current rotation and translation matrix of the vehicle-mounted camera based on the local odometer mapping principle and the P3P pose estimation algorithm specifically includes: Construct the actual three-dimensional space of the vehicle driving scene based on the principle of local odometry mapping; Projecting the key feature points and the matching feature points that are matched to each other into the actual three-dimensional space to generate key three-dimensional points and matching three-dimensional points respectively; and The current rotation and translation matrix of the vehicle-mounted camera is calculated based on the P3P pose estimation algorithm using three pairs of non-collinear and mutually matching key three-dimensional points and matching three-dimensional points.
5. The method for calculating the posture of a vehicle-mounted camera according to claim 4, wherein: The method further comprises: Calculating actual projection errors of the remaining mutually matching key three-dimensional points and matching three-dimensional points corresponding to the rotation and translation matrices based on a random sampling algorithm model; and The minimum value of the actual projection error is calculated based on the error optimization algorithm model, and the rotation and translation matrix is recalculated and updated according to the key three-dimensional points and matching three-dimensional points that meet the minimum value and match each other.
6. The method for calculating the posture of a vehicle-mounted camera according to claim 1, wherein: The region of interest is determined by judging whether the actual number of corner points in each region of the reference scene image is greater than a predetermined number.
7. A vehicle-mounted camera posture calculation device connected to a vehicle-mounted camera of a motor vehicle, characterized in that: The vehicle-mounted camera posture calculation device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the vehicle-mounted camera posture calculation method according to any one of claims 1 to 6.
8. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the vehicle-mounted camera posture calculation method according to any one of claims 1 to 6.