Binocular vision ranging method, system and equipment based on unreal engine
By constructing a virtual city using Unreal Engine for binocular ranging and employing machine learning models for error correction, the problems of high cost and low accuracy in ranging in high-altitude urban environments have been solved, achieving high-precision and robust ranging results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-26
- Publication Date
- 2026-04-10
AI Technical Summary
Traditional binocular ranging suffers from problems such as high data acquisition costs, large calibration errors, lack of high-altitude characteristics in datasets, and lack of support from simulation platforms in high-altitude urban environments, making it difficult to achieve accurate and robust ranging.
A virtual city is built using Unreal Engine, image data is collected and calibrated, machine learning models are used for error correction, and virtual ground truth is combined to train the model to improve ranging accuracy.
It significantly reduces data acquisition costs, improves ranging accuracy and robustness, adapts to complex environmental conditions, achieves reproducibility and comparability of results, and enhances real-world ranging accuracy.
Smart Images
Figure CN121837337A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of distance measurement, in particular to a binocular vision distance measurement method, system and device based on Unreal Engine. BACKGROUND
[0002] Traditional binocular distance measurement calibrates cameras in real environment through calibration board such as checkerboard (Zhang Zhengyou method, etc.), collects synchronous images of left and right cameras, and obtains depth after relevant processing. This method is commonly used in indoor experiments or outdoor road scenes. However, when binocular distance measurement is used in plateau cities, the following technical defects exist: 1. High real collection cost and uncontrollable scene: Large-scale data collection in plateau cities has high cost (manpower, equipment, climate restrictions), and it is difficult to reproduce strictly consistent multiple environmental conditions for algorithm comparison test.
[0003] 2. Calibration error and real-time calibration in dynamic scene are difficult to guarantee: Traditional offline calibration method represented by checkerboard is difficult to be applied in real time in dynamic unmanned aerial vehicle flight scene, and manual calibration brings system error.
[0004] 3. Existing data sets and simulation scenes lack plateau features and complexity of urban scenes: Public data sets are mostly biased towards road / autonomous driving or laboratory samples, and cannot fully cover phenomena such as special light, thin atmosphere and strong reflection (snow) in plateau; general simulation platforms do not have special support for modeling of plateau environmental features. SUMMARY
[0005] The purpose of the present application is to provide a binocular vision distance measurement method, system and device based on Unreal Engine, which can improve the accuracy and robustness of binocular distance measurement in complex environments such as plateau cities.
[0006] To achieve the above purpose, the present application provides the following solutions: In a first aspect, the present application provides a binocular vision distance measurement method based on Unreal Engine, comprising: constructing a virtual city through Unreal Engine; the virtual city is used to control virtual weather according to instructions; configuring a virtual binocular camera in the virtual city and collecting image data, calibrating the collected binocular images to obtain calibration results; wherein the binocular images include left images and right images; correcting the binocular images according to the calibration results to obtain corrected binocular images; calculating a disparity map based on the corrected binocular images and the calibration results using a stereo matching algorithm; calculating depth based on the calibration results and the disparity map to determine a three-dimensional point cloud dataset; The scene ground truth in the virtual city and corresponding three-dimensional points in the three-dimensional point cloud dataset are taken as one sample, and a plurality of samples are used to train a machine learning model to obtain an error correction model; The three-dimensional points determined in the actual city are input into the error correction model to obtain a corrected ground truth, and a distance value is calculated according to the corrected ground truth.
[0007] In a second aspect, the present application provides a binocular vision ranging system based on Unreal Engine, comprising: A virtual scene construction module is configured to construct a virtual city through Unreal Engine; the virtual city is configured to regulate virtual weather according to instructions; A virtual data acquisition module is configured to configure a virtual binocular camera in the virtual city and acquire image data, calibrate the acquired binocular image to obtain a calibration result; wherein the binocular image comprises a left image and a right image; An image calibration module is configured to correct the binocular image according to the calibration result to obtain a corrected binocular image; A parallax calculation module is configured to calculate a parallax map by using a stereo matching algorithm based on the corrected binocular image and the calibration result; A point cloud determination module is configured to calculate a depth based on the calibration result and the parallax map to determine a three-dimensional point cloud dataset; A model construction module is configured to take the scene ground truth in the virtual city and corresponding three-dimensional points in the three-dimensional point cloud dataset as one sample, and then use a plurality of samples to train a machine learning model to obtain an error correction model; A model application and distance calculation module is configured to input the three-dimensional points determined in the actual city into the error correction model to obtain a corrected ground truth, and calculate a distance value according to the corrected ground truth.
[0008] In a third aspect, the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a binocular vision ranging method based on Unreal Engine.
[0009] According to the specific embodiments provided in the application, the application discloses the following technical effects: the application builds a virtual city through the Unreal Engine, which can be a plateau city or any other required city, and is a controllable virtual scene. Through this setting, the data collection cost is significantly reduced, a variety of extreme environmental conditions can be accurately reproduced, the result reproducibility and comparability are realized, and long-term experiments and algorithm iteration are facilitated. The application also performs binocular image collection, calibration and correction based on the virtual city, reduces the manual calibration error, improves the calibration accuracy in a dynamic scene, and ensures the reliability of the depth calculation basic parameters. In addition, the application also sets an error correction model of virtual-real coupling, and transfers the advantages of virtual true value to the real system through training the model, so that the real ranging accuracy is improved, especially in the scene where the real system is difficult to annotate or collect true value.
[0010] The application enables low-cost, reproducible and high-confidence verification and optimization of binocular ranging algorithms in limited complex environmental conditions, and provides an error correction / adaptive parameter adjustment method for a real system, thereby significantly improving the accuracy and robustness of binocular ranging in a plateau city complex environment. BRIEF DESCRIPTION OF DRAWINGS
[0011] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.
[0012] Figure 1 A flowchart of a binocular vision ranging method based on the Unreal Engine in an embodiment of the application.
[0013] Figure 2 A structural schematic diagram of a computer device provided in an embodiment of the application. DETAILED DESCRIPTION
[0014] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.
[0015] In order to make the purpose, features and advantages of the application more obvious and easy to understand, the application will be further described in detail below with reference to the drawings and specific embodiments.
[0016] In one exemplary embodiment, asFigure 1 As shown, a binocular vision ranging method based on Unreal Engine is provided, which is executed by a computer device, specifically, can be executed by a terminal or a server, or can be executed by a terminal and a server together. In the embodiments of the present application, the following steps 101 to 107 are included.
[0017] Step 101, constructing a virtual city through Unreal Engine; the virtual city is used to control virtual weather according to instructions. The virtual city is a virtual scene, which is a programmable three-dimensional city environment based on three-dimensional modeling and rendering engine (for example, Unreal Engine 5 / Unreal Engine 5, UE5), including terrain, buildings, roads, materials, lighting and weather effects, etc.
[0018] In a specific application, the virtual city is constructed through Unreal Engine, including: inputting preset geographical features, preset weather configuration parameters and preset materials and reflection parameters into Unreal Engine UE5 for scene rendering to obtain the virtual city; wherein the preset geographical features include DEM / DSM terrain data, building models, roads and landmarks; the weather types corresponding to the preset weather configuration parameters include at least sunny, rain and snow, and lightning, which can be input into Unreal Engine in the form of a weather configuration parameter table to generate corresponding weather-related instructions.
[0019] Specifically, the terrain and key landmarks are imported and restored in UE5, the materials and reflection parameters are configured, and then the scene configuration is saved and a scene description file is generated, that is, the virtual city is obtained.
[0020] Step 102, configuring a virtual binocular camera in the virtual city and collecting image data, and calibrating the collected binocular image to obtain a calibration result; wherein the binocular image includes a left image and a right image.
[0021] The virtual camera (or virtual binocular camera) is a camera model simulated by the rendering engine in the virtual scene, which has settable internal and external parameters (such as pose including position and orientation), and is used to generate a simulated image sequence. The camera internal parameters generally include focal length f, principal point, pixel size, etc.; the external parameters refer to the pose (rotation matrix R and translation vector t) of the camera in the world coordinate system. In addition, calibration refers to the process of estimating or confirming the camera internal and external parameters.
[0022] Specifically, step 102 includes: (21) creating at least one virtual camera group in the virtual city. In actual application, related technical personnel can create left and right two (or more) virtual cameras in the scene according to needs.
[0023] (22) Set a synchronous trigger strategy, binocular baseline and camera parameters for the left camera and the right camera in the virtual camera group. The camera parameters include resolution, etc., and an interface is set to export internal and external parameters, which is an automatic calibration channel and is automatically calculated by Unreal Engine.
[0024] (23) Set a UAV flight path or a static sampling point in the virtual city, and trigger the left camera and the right camera to synchronously collect left images and right images to obtain binocular images. When triggered for a UAV flight path, a sequence of binocular images collected according to timestamps is obtained; when triggered for a static sampling point, a static image frame is obtained. In the image collection process, the left and right images and the timestamps need to be synchronized to support UAV trajectory sampling and camera synchronization. Thus, the left and right image pairs and the pose / timestamp records can be obtained.
[0025] (24) Calibrate the binocular images to calculate monocular camera internal parameters and binocular camera external parameters, and use them as calibration results.
[0026] In the present application, two calibration methods can be used. The first one is automatic calibration by directly reading parameters from a virtual engine, that is, directly reading the internal and external parameters of the virtual camera in the rendering engine. The second one is a classic calibration based on images, that is, using a calibration image such as a checkerboard, and using a standard calibration algorithm (for example, Zhang method) to be compatible with the real calibration process; the specific process is as follows: after fixing the camera position, before distance measurement of all subsequent pictures, first shoot a set of checkerboard pictures, load them into Matlab software, and calibrate them by using a binocular calibration toolbox to obtain camera internal and external parameters. Then, when collecting other pictures for distance measurement, the binocular calibration step is no longer needed.
[0027] Step 103, correct the binocular images according to the calibration results to obtain corrected binocular images.
[0028] Step 104, based on the corrected binocular images and the calibration results, calculate a disparity map by using a stereo matching algorithm. The stereo matching algorithm can be a block matching (BM) algorithm, a semi-global matching (SGM) algorithm or a network model based on deep learning, such as PSMNet, GANet, etc.
[0029] Stereo rectification is to map the images of the left and right cameras to the same row and column of projections, so that the same world point is located in the same scanning row in the two images, which facilitates disparity matching. Disparity map is the difference between the horizontal coordinates of the same scene point in the left and right images after stereo rectification, which is usually represented by d(u, v). Disparity confidence is a value for measuring the reliability of disparity estimation, which can be obtained from the cost function value, cost difference, uniqueness ratio or cross-consistency.
[0030] In practical applications, through the processing of the stereo matching algorithm, the disparity map d(u, v) and the corresponding disparity confidence map C_d(u, v) can be obtained. Specifically, the depth confidence evaluation algorithm is used in the present application, the left / right disparity map output by the SGM semi-global matching algorithm, the core intermediate data structure of the SGM semi-global matching algorithm constructed in the calculation process, the cost cube, and the disparity map are used as inputs, and the corresponding output is: a single channel confidence map with the same size as the input. For each pixel position in the input, the corresponding position in the confidence map stores a single numerical value, which represents the reliability of the depth value calculated by the stereo matching algorithm at this pixel point. The disparity confidence can be sent to step 106, and when the error correction is performed, the correction is performed according to the confidence. For pixels with low confidence, the original depth value is not trusted, and the position is not corrected. If the confidence is high but there is still an error, it may be a systematic deviation, which needs to be corrected.
[0031] Step 105, calculating the depth based on the calibration result and the disparity map to determine the three-dimensional point cloud data set. Wherein, the depth / distance refers to the actual distance from the scene point to the camera. Specifically, step 105 includes: The depth map is calculated using the following formula: .
[0032] Wherein, Z(u, v) is the depth value corresponding to the pixel (u, v), unit: meter; f is the focal length of the monocular camera parameter, in pixels, or in a consistent unit after conversion with pixel size; B is the binocular base in the monocular camera parameter, which is the linear distance between the optical centers of the two cameras in the binocular system, which is an important parameter for depth calculation, unit: meter; d(u, v) is the disparity value corresponding to the pixel (u, v) in the disparity map, unit: pixel.
[0033] The pixel coordinates are back-projected to the three-dimensional coordinates (X, Y, Z) in the camera coordinate system using the following formula to obtain the three-dimensional point: .
[0034] Wherein, c x , c y are the principal point coordinates in the monocular camera parameter.
[0035] In addition, the present application also calculates the uncertainty propagation (depth uncertainty estimation): if the standard deviation of the disparity is σ d , the uncertainty of the depth can be approximately expressed as: The formula is used to evaluate the depth confidence at different distances / disparities.
[0036] Through the above processing, an initial point cloud set can be obtained, and on this basis, median filtering, hole filling and noise removal processing are performed to obtain a final three-dimensional point cloud data set.
[0037] In step 106, the ground truth in the virtual city is compared with the corresponding three-dimensional point in the three-dimensional point cloud data set as a sample, and then a plurality of samples are used to train a machine learning model to obtain an error correction model.
[0038] Virtual-real coupling refers to a process of comparing, correcting and training real acquisition data or algorithm results with available ground truth data in a virtual scene. In the present application, step 106 includes: (61) The ground truth in the virtual city is compared with the three-dimensional point in the three-dimensional point cloud data set pixel by pixel, and an error field is calculated. The error types in the error field include absolute error, relative error and mean square error.
[0039] (62) The ground truth Z_true in the virtual city is compared with the corresponding three-dimensional point Z_est in the three-dimensional point cloud data set as a sample, a regression model is trained based on a plurality of samples, and the model is optimized (specifically, whether the model is effective is verified) using the error field to obtain an error correction model. The regression model used can also be replaced by a lightweight neural network or a lookup table, etc. The model can extract a feature vector F including disparity confidence, local texture entropy, brightness statistics, camera height, lighting condition coding, etc. according to pixels / local blocks, and then train the model g: F -> deltaZ (or directly predict a scaling factor s, so that Z_corrected = s·Z_est), and finally apply the model g to real acquisition data for error correction.
[0040] Robustness testing refers to a test process for verifying the stability of algorithm performance under multiple environmental conditions (such as different lightings, weather). Correspondingly, after obtaining the error correction model, the present application further includes: defining a plurality of test scenes, each of which includes time information, weather information, visibility and noise information; executing any of the test scenes on the virtual city, automatically performing data acquisition, ranging, error statistics and other processing, constructing an error correction model and calculating corresponding performance data, which can be a performance curve in the ROC / PR style and an error distribution table; generating a visual report according to all the error correction models and corresponding performance data. In addition, according to the visual report, model selection and algorithm improvement can be performed as needed.
[0041] In step 107, the three-dimensional points determined in the actual city are input into the error correction model to obtain corrected ground truth, and then the distance value is calculated according to the corrected ground truth.
[0042] In practical applications, the application can use public simulation / synthetic datasets such as Middlebury, KITTI, ETH3D, etc. to evaluate the stereo matching algorithm.
[0043] Based on the same inventive concept, the application also provides a system. The system provides a solution to the problem similar to the implementation scheme described in the above method, so the specific limitations in one or more system embodiments provided below can refer to the limitations of the method described above, and will not be repeated here.
[0044] In one exemplary embodiment, a Unreal Engine-based binocular vision ranging system is provided, comprising: A virtual scene construction module is configured to construct a virtual city by using a Unreal Engine; the virtual city is configured to regulate virtual weather according to instructions.
[0045] A virtual data acquisition module is configured to configure a virtual binocular camera in the virtual city and acquire image data, calibrate the acquired binocular image to obtain a calibration result; wherein the binocular image comprises a left image and a right image.
[0046] An image calibration module is configured to correct the binocular image according to the calibration result to obtain a corrected binocular image.
[0047] A parallax calculation module is configured to calculate a parallax map based on the corrected binocular image and the calibration result by using a stereo matching algorithm.
[0048] A point cloud determination module is configured to calculate a depth based on the calibration result and the parallax map to determine a three-dimensional point cloud dataset.
[0049] A model construction module is configured to use a scene ground truth in the virtual city and a corresponding three-dimensional point in the three-dimensional point cloud dataset as a sample, and then train a machine learning model using a plurality of samples to obtain an error correction model.
[0050] A model application and distance calculation module is configured to input a three-dimensional point determined in an actual city into the error correction model to obtain a corrected ground truth, and then calculate a distance value according to the corrected ground truth.
[0051] In summary, the application uses UE5 to construct a controllable highland virtual scene and perform batch collection and testing, which can significantly reduce data collection costs, accurately reproduce a variety of extreme environmental conditions, realize result reproducibility and comparability, and facilitate long-term experiments and algorithm iteration. The application provides automatic calibration of virtual engine export parameters and online fine-tuning of real scenes, which can reduce manual calibration errors, improve calibration accuracy in dynamic scenes, and ensure the reliability of depth calculation basic parameters. The virtual city constructed by the application contains weather information, which can enhance the adaptability of the algorithm in the highland city environment and improve the success rate of model migration to the real highland environment. The application also designs a virtual-real coupling error correction mechanism, which transfers the advantages of virtual true values to real systems through training correction models, so that the real ranging accuracy is improved, especially in real scenes where it is difficult to annotate or collect true values.
[0052] In an exemplary embodiment, a computer device, which can be a server or a terminal, is provided, and an internal structure diagram of the computer device can be as shown in Figure 2 The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through network connection. The computer program is executed by the processor to implement the binocular vision ranging method based on the Unreal Engine.
[0053] Those skilled in the art can understand that Figure 2 The structure shown in the above
[0054] In an exemplary embodiment, a computer device is also provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the steps in the above method embodiments.
[0055] In an exemplary embodiment, a computer readable storage medium storing a computer program is provided, the computer program, when executed by a processor, implements the steps of any of the above method embodiments.
[0056] In an exemplary embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the steps of any of the above method embodiments.
[0057] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.
[0058] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. Any reference to memory, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.
[0059] The database involved in each of the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, and the like, without being limited thereto. The processor involved in each of the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, and the like, without being limited thereto.
[0060] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but it should be considered that any combination of the technical features is within the scope of the present disclosure, as long as there is no contradiction.
[0061] The principles and implementation manners of the present application are described by using specific examples herein, and the above embodiments are only used to help understand the method of the present application and its core idea. Meanwhile, for those skilled in the art, the specific implementation manners and application ranges can be changed according to the idea of the present application. In summary, the content of the present description should not be understood as a limitation of the present application.
Claims
1. A binocular vision ranging method based on Unreal Engine, characterized in that, The method includes: A virtual city is constructed using Unreal Engine; this virtual city is used to control virtual weather according to instructions. A virtual binocular camera is configured in the virtual city to acquire image data. The acquired binocular images are calibrated to obtain calibration results. The binocular images include a left image and a right image. The binocular image is corrected based on the calibration results to obtain a corrected binocular image; Based on the corrected binocular images and the calibration results, a stereo matching algorithm is used to calculate the disparity map; The depth is calculated based on the calibration results and the disparity map to determine the 3D point cloud dataset. The scene ground truth in the virtual city and the corresponding 3D points in the 3D point cloud dataset are used as a sample. Then, multiple samples are used to train a machine learning model to obtain an error correction model. The three-dimensional points determined in the actual city are input into the error correction model to obtain the corrected true value, and then the distance value is calculated based on the corrected true value.
2. The binocular vision ranging method based on Unreal Engine according to claim 1, characterized in that, Building virtual cities using Unreal Engine includes: Preset geographical features, preset meteorological configuration parameters, and preset material and reflection parameters are input into Unreal Engine UE5 for scene rendering to obtain a virtual city; wherein, the preset geographical features include DEM / DSM terrain data, building models, roads and landmarks; the weather types corresponding to the preset meteorological configuration parameters include at least sunny, rainy / snowy, and thunderstorms.
3. The binocular vision ranging method based on Unreal Engine according to claim 1, characterized in that, A virtual stereo camera is configured in the virtual city to acquire image data. The acquired stereo images are then calibrated to obtain calibration results, including: Create at least one virtual camera group in the virtual city; Set the synchronization triggering strategy, binocular baseline and camera parameters for the left and right cameras in the virtual camera group; In the virtual city, a drone flight path or static sampling point is set, and the left camera and the right camera are triggered to simultaneously acquire left and right images to obtain binocular images; The binocular images are calibrated to calculate the intrinsic parameters of the monocular camera and the extrinsic parameters of the binocular camera, and the results are used as calibration results.
4. The binocular vision ranging method based on Unreal Engine according to claim 1, characterized in that, The stereo matching algorithm is either the block matching BM algorithm, the semi-global matching SGM algorithm, or a deep learning-based network model.
5. The binocular vision ranging method based on Unreal Engine according to claim 1, characterized in that, Based on the calibration results and the disparity map, the depth is calculated to determine the 3D point cloud dataset, including: The depth map is calculated using the following formula: ； Where Z(u,v) is the depth value corresponding to pixel (u,v), f is the camera focal length in the monocular camera intrinsic parameters; B is the binocular baseline in the monocular camera intrinsic parameters; and d(u,v) is the disparity value corresponding to pixel (u,v) in the disparity map. The pixel coordinates are back-projected to the 3D coordinates (X, Y, Z) in the camera coordinate system using the following formula to obtain the 3D point: ； Among them, c x c y These are the coordinates of the principal point in the intrinsic parameters of a monocular camera.
6. The binocular vision ranging method based on Unreal Engine according to claim 1, characterized in that, The scene ground truth in the virtual city and the corresponding 3D points in the 3D point cloud dataset are used as a sample. Then, multiple samples are used to train a machine learning model to obtain an error correction model, including: The scene ground truth in the virtual city is compared pixel by pixel with the 3D points in the 3D point cloud dataset, and an error field is calculated; the error types in the error field include absolute error, relative error and mean square error. The scene ground truth in the virtual city and the corresponding 3D points in the 3D point cloud dataset are taken as a sample. A regression model is trained based on multiple samples, and the error field is used to optimize the model to obtain an error correction model.
7. The binocular vision ranging method based on Unreal Engine according to claim 1, characterized in that, After obtaining the error correction model, the method further includes: Multiple test scenarios are defined; each test scenario includes time information, weather information, visibility and noise information; Perform any of the aforementioned test scenarios on the virtual city, then construct an error correction model and calculate the corresponding performance data; Based on all the error correction models and their corresponding performance data, generate a visualization report.
8. A binocular vision ranging system based on Unreal Engine, characterized in that, The system includes: The virtual scene construction module is used to build a virtual city using the Unreal Engine; the virtual city is used to control virtual weather according to instructions. The virtual data acquisition module is used to configure a virtual binocular camera in the virtual city and acquire image data, and to calibrate the acquired binocular images to obtain calibration results; wherein, the binocular images include a left image and a right image; An image calibration module is used to correct the binocular image based on the calibration result to obtain a corrected binocular image; The disparity calculation module is used to calculate the disparity map based on the corrected binocular image and the calibration result using a stereo matching algorithm; The point cloud determination module is used to calculate the depth based on the calibration results and the disparity map to determine the three-dimensional point cloud dataset. The model building module is used to take the scene ground truth in the virtual city and the corresponding 3D points in the 3D point cloud dataset as a sample, and then use multiple samples to train a machine learning model to obtain an error correction model. The model application and distance calculation module is used to input the three-dimensional points determined in the actual city into the error correction model to obtain the corrected true value, and then calculate the distance value based on the corrected true value.
9. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the binocular vision ranging method based on Unreal Engine according to any one of claims 1-7.