Immersive rail transit simulation training system and training method based on virtual reality

Abstract

The present invention discloses an immersive rail transit simulation training system and training method based on virtual reality, which relates to the field of rail transit simulation training technology, and includes the following steps: in a rail transit simulation training environment, the trainee's current virtual driving perspective is continuously monitored, and the visual parameter information of the scene in front of the trainee is obtained in real time. The present invention solves the lag problem of LOD adjustment in high-speed dynamic scenes, and improves the trainee's emergency response ability and training realism. Through real-time monitoring of the trainee's perspective, data preprocessing and index management, an efficient data set is constructed, and key indicators are extracted to quantify changes in the scene ahead. Combined with machine learning prediction, it intelligently identifies obstacles approaching at high speed, adopts progressive LOD adjustment, initially combines low-precision LOD + motion blur to reduce the computational burden, gradually improves LOD, ensures visual continuity and computing performance optimization, and improves the authenticity, stability and training effectiveness of the simulation system.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit simulation training, and in particular to an immersive rail transit simulation training system and a training method based on virtual reality. Background Art

[0002] The virtual reality-based immersive rail transit simulation training system is a professional training platform that integrates VR (virtual reality), 3D modeling, real-time interaction, physical simulation, and intelligent assessment technologies. It aims to provide a realistic training environment for rail transit operations, dispatching, maintenance, and emergency response. The system utilizes high-precision 3D modeling to recreate real-world rail transit scenarios, including trains, tracks, stations, and signaling systems. It also incorporates force feedback, voice interaction, and behavior tracking technologies to enable highly immersive operational training for trainees in a virtual environment. The system can simulate a variety of operational conditions, such as normal driving, troubleshooting, and emergency response. Through data recording and AI-powered intelligent analysis, it conducts real-time assessments of trainee operations and provides personalized training plans. This helps enhance rail transit practitioners' emergency response capabilities and operational proficiency, improves training efficiency, and reduces actual operational risks.

[0003] The existing technology has the following deficiencies: The optimized expression:

[0004] Existing technologies typically dynamically adjust the Level of Detail (LOD) based on the changing area in front of the trainee's viewpoint. The core principle is to prioritize rendering of key areas within the trainee's current viewpoint, allocating less computing resources to less-interested areas. This improves rendering efficiency, optimizes frame rates, reduces computational load, and ensures immersion. However, if an obstacle suddenly appears ahead (such as a flock of birds or falling rocks) and threatens to collide with the train while it's in motion, existing technologies often fail to quickly and adaptively adjust the LOD. This can lead to training distortion, decreased system performance, and safety response errors, compromising the effectiveness and immersiveness of simulation training.

[0005] Due to the lag in LOD adjustment, distant obstacles retain low precision, making it difficult for trainees to accurately judge their size, shape, and motion. For example, distant flocks of birds or falling rocks may appear blurry, fragmented, or have unclear outlines due to the low LOD level, making it difficult for trainees to quickly identify whether they pose a threat. Only when a train approaches does the LOD suddenly switch to high precision, causing a sudden change in vision, affecting perceptual stability, and disrupting trainees' reaction time, thus hampering their ability to respond to emergencies.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0007] The purpose of the present invention is to provide an immersive rail transit simulation training system and training method based on virtual reality, which solves the lag problem of LOD adjustment in high-speed dynamic scenes and improves the students' emergency response capabilities and training realism. By real-time monitoring of students' perspectives, data preprocessing and index management, an efficient data set is constructed, and key indicators are extracted to quantify changes in the scene ahead. Combined with machine learning predictions, high-speed approaching obstacles are intelligently identified, and progressive LOD adjustment is adopted. In the initial stage, low-precision LOD + motion blur is combined to reduce the computational burden, and LOD is gradually improved to ensure visual continuity and optimization of computing performance, thereby improving the authenticity, stability and training effectiveness of the simulation system to solve the problems in the above-mentioned background technology.

[0008] In order to achieve the above-mentioned object, the present invention provides the following technical solution: an immersive rail transit simulation training method based on virtual reality, comprising the following steps:

[0009] In the rail transit simulation training environment, the trainee's current virtual driving perspective is continuously monitored to obtain the visual parameter information of the scene in front of the trainee in real time;

[0010] After preprocessing the acquired visual parameter information, the preprocessed visual data is stored, indexed and managed to form a data set that can be quickly read and analyzed, further compressing the complex image data into quantitative and analytical indicators;

[0011] Extract key indicators reflecting the rapid movement of objects in the student's forward perspective from the data set, and conduct in-depth analysis and processing of the extracted indicators to quantify the speed of change of the scene ahead, thereby improving the dynamic environment perception ability and providing a reliable basis for subsequent prediction and LOD adaptive adjustment;

[0012] The quantified key indicators are used as input features and fed into a pre-trained machine learning model. The model intelligently analyzes the dynamic changes in the scene in front of the trainee and accurately assesses whether there are obstacles approaching at high speed.

[0013] When an obstacle is detected to be about to appear quickly in front of the trainee, the system automatically prioritizes increasing the level of detail of the obstacle area. In the initial stage of the obstacle approaching at high speed, low-precision LOD combined with motion blur is used to maintain reasonable visual perception and avoid consumption of computing resources. As the obstacle approaches the train, it gradually and smoothly transitions to high-precision LOD to ensure the continuity of the visual effect, while optimizing computing performance and avoiding visual mutations caused by LOD mutations.

[0014] Preferably, after preprocessing the acquired visual parameter information, the processed visual data is stored, indexed, and managed to form a data set that can be quickly read and analyzed, and the complex image data is further compressed into quantitative and analyzable indicators. The specific steps are as follows:

[0015] First, the pre-processed data is structured and stored in an efficient data format to facilitate subsequent efficient query and calculation.

[0016] Next, an indexing system is established for the data set to ensure fast retrieval of data in time series.

[0017] Subsequently, dimensionality reduction methods are applied to reduce data complexity, increase processing speed, and optimize storage resource usage;

[0018] Finally, through real-time data stream processing, the data set is dynamically updated to ensure that the latest analysis results are always stored and can quickly respond to subsequent prediction tasks, improving the real-time and intelligent level of LOD adjustment.

[0019] Preferably, key indicators reflecting the rapid movement of objects in the student's front perspective are extracted from the data set, and the extracted indicators include the degree of perspective change of fixed objects in the background and the rate of change of the area of ​​the projection of the front obstacle on the ground. The degree of perspective change of fixed objects in the background and the rate of change of the area of ​​the projection of the front obstacle on the ground are further analyzed and processed under the detection window to generate a perspective contraction reference value and a shadow stretch reference value respectively. The perspective contraction reference value is used to quantify the degree of perspective change of fixed objects in the background, and the shadow stretch reference value quantifies the rate of change of the projection area of ​​the front obstacle on the ground. The dynamic change speed of the front scene is quantified by the perspective contraction reference value and the shadow stretch reference value, thereby enhancing the perception ability of the high-speed motion environment and providing reliable data support for subsequent obstacle prediction and LOD adaptive adjustment.

[0020] Preferably, the specific steps of further analyzing and processing the perspective change degree of the fixed object in the background in the detection window to generate the perspective contraction reference value are as follows:

[0021] First, multiple reference background points are selected within the detection window. Then, the perspective shrinkage rate of the reference background points is calculated by changing the projection ratio of consecutive frames. The calculation expression is as follows:

[0022] ;

[0023] Where, is the background perspective shrinkage rate, is the current height of the background object, is the initial height of the background object, is the distance of the fixed background object in front of the student, is the viewing angle of the background object, is the focal length;

[0024] The perspective contraction of a single background object will be affected by noise and local scene changes. Therefore, the data of multiple background points are integrated to improve the stability and robustness of the perspective contraction reference value. The calculation expression of the perspective contraction reference value is as follows:

[0025] ;

[0026] Where, is the perspective contraction reference value, It is The perspective shrinkage rate of a fixed background object, is an exponential decay weight, is the natural base, is the adjustment coefficient, It is The actual distance between the background fixed object and the student's position, is the total number of background fixed objects.

[0027] Preferably, the specific steps of further analyzing and processing the area change rate of the projection of the front obstacle on the ground in the detection window to generate the shadow extension reference value are as follows:

[0028] First, within the detection window, the continuous image frames of the obstacle in front are analyzed, and the size of the obstacle's projection area on the ground is extracted through image processing methods. The expression is as follows:

[0029] ;

[0030] Where, is the area change trend index, is the area of ​​the last frame of shadow in the detection window, is the area of ​​the first frame shadow in the detection window, The first The area of ​​the frame shadow, is the maximum change in shadow area between consecutive frames within the detection window, is the minimum change in shadow area between consecutive frames within the detection window, is a very small stability constant, is the total number of frames in the detection window;

[0031] In area change trend index On this basis, the area trend characteristics are further converted into shadow extension reference values, and the calculation expression is as follows:

[0032] ;

[0033] Where, is the shadow stretch reference value, is the maximum shadow area observed in the current detection window, is the shadow area at the initial stage of the detection window.

[0034] Preferably, the quantified perspective contraction reference value and shadow stretch reference value are used as input features and input into a pre-trained machine learning model, and a rapid approach risk coefficient is generated by the machine learning model. The rapid approach risk coefficient is used to evaluate whether there is a rapidly approaching obstacle in front of the student.

[0035] Preferably, a rapid approach risk coefficient generated when evaluating whether there is a rapidly approaching obstacle in front of the trainee using a pre-trained machine learning model is compared with a pre-set rapid approach risk coefficient reference threshold to identify whether there is a rapidly approaching obstacle in front of the trainee. The specific steps are as follows:

[0036] If the rapid approach risk coefficient is greater than the preset rapid approach risk coefficient reference threshold, it indicates that an obstacle will soon appear in front of the current student; if the rapid approach risk coefficient is less than or equal to the preset rapid approach risk coefficient reference threshold, it indicates that there will be no obstacle in front of the current student.

[0037] Preferably, when an obstacle is identified to appear quickly in front of the trainee, the specific steps of automatically and preferentially increasing the level of detail of the obstacle area are as follows:

[0038] When an obstacle is detected in front of the trainee, the trainee will be With the rapid approach to the risk factor reference threshold, the LOD is dynamically adjusted to pre-optimize the LOD level to ensure that students can perceive obstacles in time while reducing the waste of computing resources. The LOD pre-improvement strategy formula is as follows:

[0039] ;

[0040] Where, It is the pre-adjusted LOD level, indicating the LOD value after preliminary adjustment. Is the basic LOD value, that is, the initial LOD setting value before adjustment, It is the LOD adjustment coefficient, which determines the extent of LOD improvement. It is rapidly approaching the reference threshold of the risk factor;

[0041] When an obstacle approaches the train at high speed, the LOD is dynamically adjusted using a progressive LOD adjustment strategy to achieve an optimal balance between visual perception and computing resource consumption. The LOD dynamic smooth transition strategy is as follows:

[0042] ;

[0043] Where, is the final LOD level, is the LOD refinement coefficient, is the natural base, is the transition steepness control coefficient, is the normalized obstacle distance, It is the LOD transition threshold, which indicates the critical point of switching from low-precision LOD to high-precision LOD.

[0044] The immersive rail transit simulation training system based on virtual reality includes a virtual driving perspective monitoring module, a visual data preprocessing and management module, a dynamic environment perception and quantitative analysis module, an intelligent obstacle detection and prediction module, and a LOD intelligent adaptive adjustment module:

[0045] The virtual driving perspective monitoring module continuously monitors the trainee's current virtual driving perspective in a rail transit simulation training environment, and obtains visual parameter information of the scene in front of the trainee in real time;

[0046] The visual data preprocessing and management module preprocesses the acquired visual parameter information, stores, indexes and manages the preprocessed visual data to form a data set that can be quickly read and analyzed, and further compresses the complex image data into quantitative and analytical indicators;

[0047] The dynamic environment perception and quantitative analysis module extracts key indicators reflecting the rapid movement of objects in the trainee's forward perspective from the data set, and conducts in-depth analysis and processing of the extracted indicators to quantify the speed of change of the scene ahead, thereby improving the dynamic environment perception ability and providing a reliable basis for subsequent prediction and LOD adaptive adjustment;

[0048] The intelligent obstacle detection and prediction module uses quantified key indicators as input features and feeds them into a pre-trained machine learning model. The model then intelligently analyzes the dynamic changes in the scene ahead of the trainee and accurately assesses whether there are obstacles approaching at high speed.

[0049] The LOD intelligent adaptive adjustment module automatically prioritizes increasing the level of detail in the obstacle area when it detects an obstacle is about to appear quickly in front of the trainee. In the initial stage of the obstacle approaching at high speed, it uses low-precision LOD combined with motion blur to maintain reasonable visual perception and avoid consumption of computing resources. As the obstacle approaches the train, it gradually and smoothly transitions to high-precision LOD to ensure the continuity of the visual effect, while optimizing computing performance and avoiding visual mutations caused by LOD mutations.

[0050] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0051] The present invention effectively solves the problem of adjustment lag of the existing LOD (Level of Detail) mechanism in high-speed dynamic scenes by introducing an immersive rail transit simulation training method based on virtual reality, thereby improving the trainees' emergency response capabilities and training realism in complex rail transit environments. By continuously monitoring the trainees' forward perspective, obtaining visual parameters in real time, and combining data preprocessing and index management, a data set that can be quickly read and analyzed is constructed, reducing redundant calculations of original image data and improving system operation efficiency. At the same time, through the extraction and quantitative analysis of key indicators, the system can accurately evaluate the dynamic changes of the scene ahead, and use pre-trained machine learning models to intelligently predict high-speed approaching obstacles, making the LOD adjustment strategy forward-looking and adaptive. When an obstacle approaches at high speed, a progressive LOD transition is adopted, initially combining low-precision LOD + motion blur to reduce the computational burden, and then smoothly improving the LOD to avoid the visual impact caused by LOD mutations and improve the consistency of the visual experience. While optimizing the allocation of computing resources, the overall solution enhances the ability to perceive high-speed dynamic environments, making the simulation training system more realistic, stable, and effective in the rail transit field, providing trainees with more accurate operational judgments and improving their ability to respond to emergencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0053] Figure 1 This is a flow chart of the method of the immersive rail transit simulation training method based on virtual reality of the present invention.

[0054] Figure 2 This is a module diagram of the immersive rail transit simulation training system based on virtual reality of the present invention. DETAILED DESCRIPTION

[0055] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0056] The present invention provides Figure 1 The immersive rail transit simulation training method based on virtual reality shown includes the following steps:

[0057] In the rail transit simulation training environment, the trainee's current virtual driving perspective is continuously monitored to obtain the visual parameter information of the scene in front of the trainee in real time;

[0058] The core goal of this step is to keep abreast of changes in the scene the trainee sees within the virtual rail transit environment. During the simulation, the system uses virtual cameras, sensors, or a VR headset to capture visual images, depth information, and other auxiliary data directly in front of the trainee. Real-time acquisition of this visual parameter information ensures the system can promptly perceive dynamic changes in the scene ahead, such as rapidly approaching objects and changes in ambient lighting, laying a solid foundation for subsequent preprocessing and obstacle detection.

[0059] After preprocessing the acquired visual parameter information, the preprocessed visual data is stored, indexed and managed to form a data set that can be quickly read and analyzed, further compressing the complex image data into quantitative and analytical indicators;

[0060] Preprocessing removes noise, filters out redundant data, and extracts key features, providing cleaner and more accurate input for subsequent data analysis and key metric extraction. First, image processing algorithms (such as Gaussian filtering and bilateral filtering) are used to remove noise and normalize brightness and depth information to ensure data consistency across multiple scenarios. Then, classic machine vision algorithms (such as Canny edge detection and background subtraction) or deep learning algorithms (such as YOLO and Mask R-CNN) are used to detect or segment objects in the image and identify possible obstacle outlines. Furthermore, if the system includes multiple sensors (such as a depth camera and a standard RGB camera), timing synchronization and calibration are required to ensure the accuracy and consistency of subsequent analysis. This preprocessing ensures clearer, more accurate, and more targeted image data, reducing errors in subsequent machine learning models and improving the accuracy of obstacle detection and LOD adjustment.

[0061] After preprocessing the acquired visual parameter information, the processed visual data is stored, indexed, and managed to form a data set that can be quickly read and analyzed, and the complex image data is further compressed into quantitative and analyzable indicators. The specific steps are as follows: First, the preprocessed data is structured and stored in an efficient data format (such as HDF5, Parquet) for subsequent efficient query and calculation; then, an index system is established for the data set, including key fields such as timestamp, scene ID, target category, position coordinates, motion vector, etc., to ensure that the data can be quickly retrieved in time series; then, dimensionality reduction methods (such as PCA, t-SNE) are applied to reduce data complexity, increase processing speed, and optimize storage resource usage; finally, through real-time data stream processing (such as Kafka, Flink), the data set is dynamically updated to ensure that the latest analysis results are always stored, and can quickly respond to subsequent prediction tasks, thereby improving the real-time and intelligent level of LOD adjustment.

[0062] Extract key indicators reflecting the rapid movement of objects in the student's forward perspective from the data set, and conduct in-depth analysis and processing of the extracted indicators to quantify the speed of change of the scene ahead, thereby improving the dynamic environment perception ability and providing a reliable basis for subsequent prediction and LOD adaptive adjustment;

[0063] Key indicators reflecting the rapid movement of objects in the trainee's forward perspective are extracted from the data set. The extracted indicators include the degree of perspective change of fixed objects in the background and the rate of change of the area of ​​the projected obstacle on the ground. These are further analyzed and processed within the detection window to generate a perspective contraction reference value and a shadow stretch reference value, respectively. The perspective contraction reference value is used to quantify the degree of perspective change of fixed objects in the background. It is used to measure the visual shrinkage rate of static structures in the scene (such as tracks, signals, platforms, and distant buildings), thereby inferring whether there are high-speed moving objects ahead or the intensity of dynamic changes in the environment. The shadow stretch reference value quantifies the rate of change of the projected area of ​​the obstacle ahead on the ground and is used to detect whether the object ahead is approaching the train at high speed. The perspective contraction reference value and the shadow stretch reference value are used to quantify the dynamic change rate of the scene ahead, thereby enhancing the perception of high-speed moving environments and providing reliable data support for subsequent obstacle prediction and LOD adaptive adjustment.

[0064] A noticeable shrinkage of the perspective of fixed objects in the background typically indicates a fast-moving object in front of the student's field of view or that the overall scene is undergoing dramatic dynamic changes. This is because, under normal circumstances, perspective changes are determined by the relative motion of the observer. When the student (or simulated camera) is traveling at a constant speed, the perspective changes of distant fixed objects are slow and smooth. However, if the background perspective shrinkage suddenly accelerates, it indicates a dramatic shift in the relative motion pattern within the field of view. This may be due to an object entering the student's field of view at high speed, causing a dynamic shift in the overall visual reference frame. For example, if a distant signal light, building, or track sign shrinks significantly over a short period of time, while new structural edges or texture features appear in the foreground, it could be that a high-speed moving object (such as a flock of birds, falling rocks, or pedestrians) has entered the student's field of view, and its movement direction is aligned with or close to that of the train. When the background perspective shrinkage reference value reaches a certain threshold, and the new contours and motion trajectories in the field of view match a high-speed motion pattern, the system can reasonably infer that an object is approaching rapidly in front of the student. Compared with recognition methods directly based on target detection, perspective contraction provides an indirect motion clue based on environmental deformation. Especially when obstacles are still blurred or far away, it can help the simulation system detect possible dynamic threats ahead earlier, improve the lead time of LOD adjustment and the dynamic adaptability of the simulation environment.

[0065] The specific steps for further analyzing and processing the perspective change degree of the fixed object in the background under the detection window to generate the perspective contraction reference value are as follows:

[0066] First, multiple reference background points are selected within the detection window, such as distant traffic lights, track signs, platform edges, and building outlines. These objects are static and unchanging in reality. Then, the perspective shrinkage rate of the reference background points is calculated by changing the projection scale of consecutive frames. The calculation expression is as follows: ;

[0067] Where, is the background perspective contraction rate, which indicates the perspective contraction speed of a fixed background object in the student's perspective, that is, the rate at which the projection height of the object in the image is reduced. Is the current height of the background object (in pixels), which refers to the height measurement value of a fixed object in the image frame. is the initial height of the background object, which refers to the height of the same object in the previous frame or the initial frame. is the distance of the fixed background object in front of the student, Is the angle of view change of the background object (in radians), that is, the angular offset of the object in the field of view in consecutive frames. is the focal length, which represents the optical focal length of the student's perspective and is used to map the relationship between actual distance and screen pixels;

[0068] By detecting the perspective changes of fixed objects in the background within the window and calculating the change in their projection height across consecutive frames, the system quantifies the rate of background perspective contraction. This is then combined with the background object's physical distance, viewing angle offset, and camera focal length to ensure the calculated results accurately reflect the perspective contraction phenomenon, providing precise data support for subsequent dynamic environment perception and obstacle detection.

[0069] The perspective contraction of a single background object will be affected by noise and local scene changes. Therefore, the data of multiple background points are integrated to improve the stability and robustness of the perspective contraction reference value. The calculation expression of the perspective contraction reference value is as follows:

[0070] ;

[0071] Where, is the perspective contraction reference value, It is The perspective shrinkage rate of a fixed background object, It is an exponential decay weight, which is used to dynamically adjust the contribution of background objects to the perspective contraction reference value, so that the perspective change of distant background has a greater impact on the overall calculation, while the impact of nearby objects is smaller. is the natural base, Is the adjustment coefficient, which controls the exponential decay rate of the weight, It is The actual distance between the background fixed object and the student's position, is the total number of background fixed objects.

[0072] By integrating perspective change information from multiple fixed background objects and assigning greater influence to distant objects through exponentially decaying weights, the system generates a perspective contraction reference value, quantifying the degree of dynamic change in the student's forward perspective. This process enhances the system's ability to perceive objects approaching at high speed from a distance, improves the accuracy of LOD adaptive adjustment, and provides reliable data support for obstacle detection and dynamic environment analysis.

[0073] The perspective contraction reference value, generated by further analyzing the degree of perspective change of fixed objects in the background within the detection window, is generated. A larger perspective contraction reference value generally indicates a fast-moving object in front of the trainee's field of view; a smaller perspective contraction reference value indicates no significant fast-moving objects within the field of view. Within the monitoring window, the calculation of the perspective contraction reference value depends on the perspective change rate of fixed objects in the background—that is, the degree and speed at which static background structures (such as tracks, traffic lights, platforms, and distant buildings) are visually reduced. When an object approaches at high speed in front of the trainee, it dynamically alters the perspective relationship in the overall field of view, causing previously stable background perspective changes to suddenly intensify, leading to a rapid increase in the perspective contraction reference value. For example, when an obstacle (such as a flock of birds or falling rocks) rapidly approaches a train from a distance, the fixed objects in the background will experience a more dramatic perspective contraction in a short period of time due to their rapid approach. This change is captured and quantified by the perspective contraction reference value. A lower perspective contraction reference value indicates a stable background perspective change, indicating no significant fast-moving objects within the field of view or that the trainee's observation environment is in a static or stable relative motion state.

[0074] A rapid increase in the rate of change in the area of ​​the obstacle's projection on the ground over a short period of time typically indicates the presence of a fast-moving object in the student's forward perspective. This is due to the direct correlation between shadow deformation and object motion. First, under natural lighting conditions, the size of an obstacle's shadow depends on the position of the light source (such as the angle of the sun) and the relative distance of the object from the ground. When the obstacle moves away from the observer, the change in the area of ​​its shadow's projection on the ground is relatively gradual. However, when the obstacle approaches at high speed, its projection increases exponentially in consecutive frames. This is because the object's physical size does not change, but its perspective magnification effect in the observer's perspective and the dynamic changes in its projection on the ground cause the shadow to lengthen rapidly.

[0075] Secondly, a sharp increase in the shadow's rate of change often indicates that the object itself is moving at a faster speed, rather than a sudden change in lighting conditions. For example, if a shadow grows from small to large in a very short period of time, it indicates that the object is moving toward the train at a high relative speed, rather than simply due to a slow change in the angle of the light source. In addition, compared to directly detecting the object's own movement, shadow changes can provide an additional motion clue when the object has not yet fully entered the student's field of view or is blocked by an object, allowing the system to predict potential dangerous objects earlier.

[0076] The specific steps for further analyzing and processing the area change rate of the projection of the obstacle in front on the ground in the detection window to generate the shadow extension reference value are as follows:

[0077] First, within the detection window, the continuous image frames of the obstacle in front are analyzed, and the size of the obstacle's projection area on the ground is extracted through image processing methods. The expression is as follows:

[0078] ;

[0079] Where, It is an area change trend index, which is used to quantify the area change trend of the obstacle's projection on the ground, helping the detection system to judge whether the shadow expands rapidly in a short period of time, and thus identify whether there is an object approaching at high speed in front of the trainee. It is the area of ​​the last shadow frame in the detection window (the latest frame), which reflects the immediate size of the current obstacle projection. It is the area of ​​the first shadow frame in the detection window (the earliest frame in the window), which reflects the scale baseline of the initially detected obstacle shadow. The first The area of ​​the frame shadow, is the maximum change in shadow area between consecutive frames within the detection window, is the minimum change in shadow area between consecutive frames within the detection window, It is a very small stability constant (such as 0.0001) used to prevent the divisor from being 0 and ensure the stable calculation of the formula. is the total number of frames in the detection window;

[0080] The purpose of the above steps is to capture the combined characteristics of the instantaneous intensity of the shadow area change and the overall change amplitude. It can not only reflect the overall scale change trend of the obstacle, but also keenly capture the drastic changes when the area suddenly increases.

[0081] In area change trend index On this basis, the area trend characteristics are further converted into shadow extension reference values, and the calculation expression is as follows:

[0082] ;

[0083] Where, is the shadow stretch reference value, It is the maximum shadow area observed in the current detection window, representing the projection scale of the obstacle when it is closest to the observer in the detection window. It is the shadow area at the initial stage of the detection window, which is used as a comparison baseline to reflect the initial state of the shadow area.

[0084] By combining logarithmic amplification with area ratio constraints, the shadow extension reference value accurately quantifies the shadow expansion rate, allowing for more sensitive detection of obstacles approaching at high speeds. This step ensures that the shadow extension reference value not only reflects the overall trend of shadow area changes but also highlights sudden increases in area. This improves the rail transit simulation system's dynamic environmental perception of fast-moving objects ahead and provides accurate data support for LOD adaptive adjustment.

[0085] The shadow stretch reference value, generated by further analyzing and processing the rate of change of the obstacle's projected area on the ground within the detection window, indicates a fast-moving object is present in the trainee's forward perspective. Conversely, a higher value indicates that no fast-moving object has been detected. This is because the rate of change of the shadow's projected area on the ground is directly related to the obstacle's motion. When an object approaches the trainee at high speed, its projected area increases dramatically in a short period of time, causing the shadow stretch reference value to rise. Conversely, when the scene ahead is relatively static or the object is moving slowly, the shadow area changes less, and the shadow stretch reference value tends to be lower.

[0086] The quantified key indicators are used as input features and fed into a pre-trained machine learning model. The model intelligently analyzes the dynamic changes in the scene in front of the trainee and accurately assesses whether there are obstacles approaching at high speed.

[0087] The quantified perspective contraction reference value and shadow stretch reference value are used as input features and input into a pre-trained machine learning model. The machine learning model generates a rapid approach risk coefficient, which is used to assess whether there is a rapidly approaching obstacle in front of the trainee.

[0088] A pre-trained machine learning model is an intelligent model that has been trained on extensive rail transit simulation data and real-world data before its official implementation, enabling efficient inference in actual operation. In this system, the model's primary task is to use input features (perspective contraction reference values ​​and shadow stretch reference values) to calculate the rapid approach risk factor and determine whether there are rapidly approaching obstacles ahead of the trainee. During the training phase, the model uses supervised or unsupervised learning methods to learn the patterns of high-speed obstacle approach using a large amount of labeled data samples. For example, the system collects and annotates the movement trajectories, size variations, and lighting characteristics of different obstacle types (such as flocks of birds, falling rocks, and pedestrians), as well as their perspective contraction and shadow stretch trends from different perspectives. Using this extensive training data, the model learns how these features relate to actual high-speed motion conditions, ultimately forming a set of optimal decision boundaries or feature mapping functions to quickly infer obstacle approach risks during actual operation.

[0089] This pre-trained machine learning model typically utilizes deep neural networks (such as CNN, LSTM, and Transformer) or traditional machine learning algorithms (such as random forests, support vector machines, and Bayesian classifiers), depending on the system's computing resources and real-time performance requirements. During the training phase, the model performs feature extraction, pattern recognition, and parameter optimization to ensure accurate judgment under different obstacle types, varying lighting conditions, and varying perspective change and shadow extension rates. Furthermore, the training process utilizes data augmentation (e.g., simulating obstacle motion in different weather conditions and at different angles), hyperparameter optimization (e.g., learning rate adjustment and regularization), and model validation (e.g., cross-validation and confusion matrix analysis) to ensure model generalization and robustness. After this pre-processing, the model does not require retraining during actual operation. Instead, it directly uses the optimized parameters for inference (prediction). Specifically, given perspective contraction and shadow extension reference values, the model can calculate the rapid approach risk factor in real time and determine whether a high-speed obstacle is approaching the student's front, providing accurate decision-making for LOD adjustments and emergency response.

[0090] The machine learning model is not limited here and can achieve the perspective contraction reference value and shadow stretch reference values Conduct comprehensive analysis to generate a quick approach to risk factors The machine learning model can be used. In order to realize the technical solution of the present invention, the present invention provides a specific implementation method;

[0091] Rapidly approaching risk factor The generation formula is as follows: , where and Perspective contraction reference values and shadow stretch reference values The preset scaling factor of and Both are greater than 0.

[0092] Preset scale factor and Represents perspective contraction reference value and shadow stretch reference values The final calculated rapid approach risk factor The role of these coefficients is to adjust the perspective contraction reference value and shadow stretch reference values In calculation to ensure that risk assessments accurately reflect reality.

[0093] Specifically:

[0094] if A larger value indicates that the system pays more attention to the rate of background perspective change, i.e. right The impact is greater and is applicable to obstacles approaching from a distance at high speed (such as a train approaching from a distance or a flock of birds flying forward from a distance).

[0095] if A larger value means that the system is more dependent on the rate of change of the ground shadow, that is, right This is more effective when there are flocks of low-flying birds or fast-moving objects on the ground (such as pedestrians and animals).

[0096] Preset scale factor and It needs to be determined through experiments, historical data analysis or machine learning optimization to adapt to the risk assessment needs in different scenarios. The calculation can maintain adaptability under different environmental conditions, obstacle types, lighting changes and other factors, thereby improving the accuracy of LOD adaptive adjustment and the reliability of emergency response.

[0097] It can be seen from the rapid approach risk coefficient that the larger the perspective contraction reference value generated after further analysis and processing of the perspective change degree of the fixed object in the background under the detection window, and the larger the shadow stretch reference value generated after further analysis and processing of the area change rate of the projection of the front obstacle on the ground under the detection window, the larger the rapid approach risk coefficient generated when evaluating whether there is a rapidly approaching obstacle in front of the student through the pre-trained machine learning model, indicating that the probability of a rapidly approaching obstacle appearing in front of the current student is greater, and vice versa, it indicates that the probability of a rapidly approaching obstacle appearing in front of the current student is smaller.

[0098] The rapid approach risk factor generated by the pre-trained machine learning model when evaluating whether there is a rapidly approaching obstacle in front of the student is compared with the pre-set rapid approach risk factor reference threshold to identify whether there is a rapidly approaching obstacle in front of the student. The specific steps are as follows:

[0099] If the rapid approach risk coefficient is greater than the preset rapid approach risk coefficient reference threshold, it indicates that an obstacle will soon appear in front of the current student; if the rapid approach risk coefficient is less than or equal to the preset rapid approach risk coefficient reference threshold, it indicates that there will be no obstacle in front of the current student.

[0100] When an obstacle is detected approaching the trainee, the system automatically prioritizes increasing the level of detail (LOD) of the obstacle area. In the initial phase of the obstacle approaching at high speed, a low-precision LOD combined with motion blur (Motion Blur) is used to maintain reasonable visual perception and avoid wasting computing resources. As the obstacle approaches the train, the system gradually and smoothly transitions to a high-precision LOD to ensure visual continuity while optimizing computing performance and avoiding visual abrupt changes caused by LOD changes.

[0101] While ensuring visual realism and immersion, optimize computing resource allocation to ensure smooth LOD (Level of Detail) adjustments, avoid sudden increases in computing load, and improve the timeliness of obstacle identification and emergency response. In rail transit simulation training, obstacles may suddenly enter the trainee's field of view and approach at high speed. If the system jumps directly from low-precision LOD to high-precision LOD, it may cause a sudden increase in computing pressure, causing lag, frame drops, and even affecting the stability of the simulation system. The sudden visual changes will also interfere with the trainee's emergency decision-making. Therefore, this step adopts a progressive LOD adjustment strategy. When an obstacle just enters the detection range, the LOD of the obstacle area is prioritized to be increased, rather than the LOD of the entire scene, to ensure the rational allocation of computing resources.

[0102] In the initial stages of an obstacle approaching at high speed, the system uses a low-precision LOD combined with motion blur (MotionBlur). This not only reduces polygon rendering computation but also enhances visual fluidity through motion blur, making the obstacle's rapid approach appear more natural, thereby improving trainees' dynamic perception and preventing distraction caused by sudden LOD changes. As the obstacle approaches the train, the system gradually increases the LOD, making details such as the obstacle's shape, texture, and shadows clearer. However, this improvement is not a sudden one, but a smooth transition based on distance or approach rate to ensure the consistency of visual changes.

[0103] This progressive LOD adjustment method can dynamically adapt to computing resources, avoiding the problem of instantaneous excessive GPU / CPU computing load caused by LOD mutations. It also reduces LOD jitter and visual jumps, making the obstacle approach process consistent with the natural observation laws of the human eye, and improving the realism and immersion of simulation training. In addition, through this method, the system can make different LOD control strategies when different types of obstacles (such as flocks of birds, falling rocks, and pedestrians) approach at high speed. For example, the LOD of large-area obstacles (such as falling rocks) can be increased earlier, while for high-speed small obstacles (such as flocks of birds), motion blur can be used to reduce computing power first, and then gradually refined rendering can be performed according to the specific trajectory. In summary, the core role of this step is to achieve intelligent LOD adjustment while taking into account computing efficiency, visual coherence, and simulation immersion, thereby improving students' emergency judgment and decision-making capabilities in high-speed dynamic scenes.

[0104] When an obstacle is detected and about to appear in front of the trainee, the specific steps for automatically prioritizing the level of detail of the obstacle area are as follows:

[0105] When an obstacle is detected in front of the trainee, the trainee will be In conjunction with the rapidly approaching risk factor reference threshold, the Level of Detail (LOD) is dynamically adjusted to pre-optimize the LOD level, ensuring that students can perceive obstacles in a timely manner while reducing computing resource waste. LOD adjustment follows the principle of gradual and smooth transition to avoid sudden visual changes caused by LOD jumps, thereby improving the immersiveness of simulation training and system response efficiency. The LOD pre-improvement strategy formula is as follows:

[0106] ;

[0107] Where, It is the pre-adjusted LOD level, indicating the LOD value after preliminary adjustment. Is the basic LOD value, that is, the initial LOD setting value before adjustment, It is the LOD adjustment coefficient, which determines the extent of LOD improvement. It is rapidly approaching the reference threshold of the risk factor;

[0108] when Less than or equal to When , the LOD adjustment amount is close to 0, and the LOD will not be adjusted to reduce the calculation load.

[0109] when Greater than , LOD is raised to the highest level to ensure that students can clearly observe obstacles.

[0110] This step uses a standardized risk factor adjustment method to ensure that the LOD increase is smoothly adjusted with the degree of danger of the approaching obstacle, avoiding sudden LOD jumps and visual mutations.

[0111] As an obstacle approaches the train at high speed, the LOD is dynamically adjusted using a progressive LOD adjustment strategy to achieve an optimal balance between visual perception and computing resource consumption. This step ensures smooth visuals and avoids visual jumps caused by LOD mutations, while providing high-precision details when necessary. The LOD dynamic smooth transition strategy is as follows:

[0112] ;

[0113] Where, Is the final LOD level, used to determine the LOD accuracy of the current frame, Is the LOD refinement coefficient, used to control the smoothness of LOD adjustment, the value is usually between , a higher value indicates a more obvious improvement in LOD accuracy, is the natural base, It is the transition steepness control coefficient, which controls the steepness of the LOD transition curve, that is, the rate of change of LOD from low precision to high precision. is the normalized obstacle distance, which indicates how close the obstacle is to the train. It is the LOD transition threshold, which indicates the critical point of switching from low-precision LOD to high-precision LOD.

[0114] The present invention effectively solves the problem of adjustment lag of the existing LOD (Level of Detail) mechanism in high-speed dynamic scenes by introducing an immersive rail transit simulation training method based on virtual reality, thereby improving the trainees' emergency response capabilities and training realism in complex rail transit environments. By continuously monitoring the trainees' forward perspective, obtaining visual parameters in real time, and combining data preprocessing and index management, a data set that can be quickly read and analyzed is constructed, reducing redundant calculations of original image data and improving system operation efficiency. At the same time, through the extraction and quantitative analysis of key indicators, the system can accurately evaluate the dynamic changes of the scene ahead, and use pre-trained machine learning models to intelligently predict high-speed approaching obstacles, making the LOD adjustment strategy forward-looking and adaptive. When an obstacle approaches at high speed, a progressive LOD transition is adopted, initially combining low-precision LOD + motion blur to reduce the computational burden, and then smoothly improving the LOD to avoid the visual impact caused by LOD mutations and improve the consistency of the visual experience. While optimizing the allocation of computing resources, the overall solution enhances the ability to perceive high-speed dynamic environments, making the simulation training system more realistic, stable, and effective in the rail transit field, providing trainees with more accurate operational judgments and improving their ability to respond to emergencies.

[0115] The present invention provides Figure 2 The immersive rail transit simulation training system based on virtual reality shown in the figure includes a virtual driving perspective monitoring module, a visual data preprocessing and management module, a dynamic environment perception and quantitative analysis module, an intelligent obstacle detection and prediction module, and a LOD intelligent adaptive adjustment module:

[0116] The virtual driving perspective monitoring module continuously monitors the trainee's current virtual driving perspective in a rail transit simulation training environment, and obtains visual parameter information of the scene in front of the trainee in real time;

[0117] The visual data preprocessing and management module preprocesses the acquired visual parameter information, stores, indexes and manages the preprocessed visual data to form a data set that can be quickly read and analyzed, and further compresses the complex image data into quantitative and analytical indicators;

[0118] The dynamic environment perception and quantitative analysis module extracts key indicators reflecting the rapid movement of objects in the trainee's forward perspective from the data set, and conducts in-depth analysis and processing of the extracted indicators to quantify the speed of change of the scene ahead, thereby improving the dynamic environment perception ability and providing a reliable basis for subsequent prediction and LOD adaptive adjustment;

[0119] The intelligent obstacle detection and prediction module uses quantified key indicators as input features and feeds them into a pre-trained machine learning model. The model then intelligently analyzes the dynamic changes in the scene ahead of the trainee and accurately assesses whether there are obstacles approaching at high speed.

[0120] The LOD intelligent adaptive adjustment module automatically prioritizes increasing the level of detail in the obstacle area when it detects an obstacle is about to appear quickly in front of the trainee. In the initial stage of the obstacle approaching at high speed, it uses low-precision LOD combined with motion blur to maintain reasonable visual perception and avoid consumption of computing resources. As the obstacle approaches the train, it gradually and smoothly transitions to high-precision LOD to ensure the continuity of the visual effect, while optimizing computing performance and avoiding visual mutations caused by LOD mutations.

[0121] The immersive rail transit simulation training method based on virtual reality provided in an embodiment of the present invention is realized by the above-mentioned immersive rail transit simulation training system based on virtual reality. The specific methods and processes of the immersive rail transit simulation training system based on virtual reality are detailed in the embodiment of the above-mentioned immersive rail transit simulation training method based on virtual reality, which will not be repeated here.

[0122] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0123] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0124] It should be noted that, in this document, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0125] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0126] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0127] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0128] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0129] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0130] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0131] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. The immersive rail transit simulation training method based on virtual reality is characterized by: The following steps are involved: In the rail transit simulation training environment, the trainee's current virtual driving perspective is continuously monitored to obtain the visual parameter information of the scene in front of the trainee in real time; After preprocessing the acquired visual parameter information, the preprocessed visual data is stored, indexed and managed to form a data set that can be quickly read and analyzed, further compressing the complex image data into quantitative and analytical indicators; Extract key indicators reflecting the rapid movement of objects in the student's forward perspective from the data set, and conduct in-depth analysis and processing of the extracted indicators to quantify the speed of change of the scene ahead, thereby improving the dynamic environment perception ability and providing a reliable basis for subsequent prediction and LOD adaptive adjustment; The quantified key indicators are used as input features and fed into a pre-trained machine learning model. The model intelligently analyzes the dynamic changes in the scene in front of the trainee and accurately assesses whether there are obstacles approaching at high speed. When an obstacle is detected approaching the trainee, the system automatically prioritizes increasing the level of detail of the obstacle area. In the initial phase of the obstacle approaching at high speed, a low-precision LOD combined with motion blur is used. As the obstacle approaches the train, the system gradually and smoothly transitions to a high-precision LOD to ensure visual continuity. Extract key indicators from the data set that reflect the rapid movement of objects in the student's front perspective. The extracted indicators include the degree of perspective change of fixed objects in the background and the rate of change of the area of ​​the projection of the front obstacle on the ground. The degree of perspective change of fixed objects in the background and the rate of change of the area of ​​the projection of the front obstacle on the ground are further analyzed and processed under the detection window to generate a perspective contraction reference value and a shadow stretch reference value, respectively. The perspective contraction reference value is used to quantify the degree of perspective change of fixed objects in the background, and the shadow stretch reference value quantifies the rate of change of the projection area of ​​the front obstacle on the ground. The perspective contraction reference value and the shadow stretch reference value are used to quantify the dynamic change speed of the front scene, thereby enhancing the perception ability of high-speed motion environments and providing reliable data support for subsequent obstacle prediction and LOD adaptive adjustment. The quantified perspective contraction reference value and shadow stretch reference value are used as input features and fed into a pre-trained machine learning model. The machine learning model then generates a rapid approach risk factor, which is used to assess whether there is a rapidly approaching obstacle in front of the trainee. The rapid approach risk factor generated by the pre-trained machine learning model when evaluating whether there is a rapidly approaching obstacle in front of the student is compared with the pre-set rapid approach risk factor reference threshold to identify whether there is a rapidly approaching obstacle in front of the student. The specific steps are as follows: If the rapid approach risk coefficient is greater than the preset rapid approach risk coefficient reference threshold, it indicates that an obstacle will soon appear in front of the current student; if the rapid approach risk coefficient is less than or equal to the preset rapid approach risk coefficient reference threshold, it indicates that there will be no obstacle in front of the current student.

2. The immersive rail transit simulation training method based on virtual reality according to claim 1 is characterized in that: After preprocessing the acquired visual parameter information, the processed visual data is stored, indexed, and managed to form a data set that can be quickly read and analyzed. The complex image data is further compressed into quantitative and analyzable indicators. The specific steps are as follows: First, the pre-processed data is structured and stored in an efficient data format to facilitate subsequent efficient query and calculation. Next, an indexing system is established for the data set to ensure fast retrieval of data in time series. Subsequently, dimensionality reduction methods are applied to reduce data complexity, increase processing speed, and optimize storage resource usage; Finally, through real-time data stream processing, the data set is dynamically updated to ensure that the latest analysis results are always stored and can quickly respond to subsequent prediction tasks, improving the real-time and intelligent level of LOD adjustment.

3. The immersive rail transit simulation training method based on virtual reality according to claim 1 is characterized in that: The specific steps for further analyzing and processing the perspective change degree of the fixed object in the background under the detection window to generate the perspective contraction reference value are as follows: First, multiple reference background points are selected within the detection window. Then, the perspective shrinkage rate of the reference background points is calculated by changing the projection ratio of consecutive frames. The calculation expression is as follows: ； Where, is the background perspective shrinkage rate, is the current height of the background object, is the initial height of the background object, is the distance of the fixed background object in front of the student, is the viewing angle of the background object, is the focal length; The perspective contraction of a single background object will be affected by noise and local scene changes. Therefore, the data of multiple background points are integrated to improve the stability and robustness of the perspective contraction reference value. The calculation expression of the perspective contraction reference value is as follows: ； Where, is the perspective contraction reference value, It is The perspective shrinkage rate of a fixed background object, is an exponential decay weight, is the natural base, is the adjustment coefficient, It is The actual distance between the background fixed object and the student's position, is the total number of background fixed objects.

4. The immersive rail transit simulation training method based on virtual reality according to claim 1 is characterized in that: The specific steps for further analyzing and processing the area change rate of the projection of the obstacle in front on the ground in the detection window to generate the shadow extension reference value are as follows: First, within the detection window, the continuous image frames of the obstacle in front are analyzed, and the size of the obstacle's projection area on the ground is extracted through image processing methods. The expression is as follows: ； Where, is the area change trend index, is the area of ​​the last frame of shadow in the detection window, is the area of ​​the first frame shadow in the detection window, The first The area of ​​the frame shadow, is the maximum change in shadow area between consecutive frames within the detection window, is the minimum change in shadow area between consecutive frames within the detection window, is a very small stability constant, is the total number of frames in the detection window; In area change trend index On this basis, the area trend characteristics are further converted into shadow extension reference values, and the calculation expression is as follows: ； Where, is the shadow stretch reference value, is the maximum shadow area observed in the current detection window, is the shadow area at the initial stage of the detection window.

5. The immersive rail transit simulation training method based on virtual reality according to claim 1 is characterized in that: When an obstacle is detected and about to appear in front of the trainee, the specific steps for automatically prioritizing the level of detail of the obstacle area are as follows: When an obstacle is detected in front of the trainee, the trainee will be With the rapid approach to the risk factor reference threshold, the LOD is dynamically adjusted to pre-optimize the LOD level to ensure that students can perceive obstacles in time while reducing the waste of computing resources. The LOD pre-improvement strategy formula is as follows: ； Where, It is the pre-adjusted LOD level, indicating the LOD value after preliminary adjustment. Is the basic LOD value, that is, the initial LOD setting value before adjustment, It is the LOD adjustment coefficient, which determines the extent of LOD improvement. It is rapidly approaching the reference threshold of the risk factor; When an obstacle approaches the train at high speed, the LOD is dynamically adjusted using a progressive LOD adjustment strategy to achieve an optimal balance between visual perception and computing resource consumption. The LOD dynamic smooth transition strategy is as follows: ； Where, is the final LOD level, is the LOD refinement coefficient, is the natural base, is the transition steepness control coefficient, is the normalized obstacle distance, It is the LOD transition threshold, which indicates the critical point of switching from low-precision LOD to high-precision LOD.

6. An immersive rail transit simulation training system based on virtual reality, used to implement the immersive rail transit simulation training method based on virtual reality as described in any one of claims 1 to 5 above, characterized in that: It includes a virtual driving perspective monitoring module, a visual data preprocessing and management module, a dynamic environment perception and quantitative analysis module, an intelligent obstacle detection and prediction module, and a LOD intelligent adaptive adjustment module: The virtual driving perspective monitoring module continuously monitors the trainee's current virtual driving perspective in a rail transit simulation training environment, and obtains visual parameter information of the scene in front of the trainee in real time; The visual data preprocessing and management module preprocesses the acquired visual parameter information, stores, indexes and manages the preprocessed visual data to form a data set that can be quickly read and analyzed, and further compresses the complex image data into quantitative and analytical indicators; The dynamic environment perception and quantitative analysis module extracts key indicators reflecting the rapid movement of objects in the trainee's forward perspective from the data set, and conducts in-depth analysis and processing of the extracted indicators to quantify the speed of change of the scene ahead, thereby improving the dynamic environment perception ability and providing a reliable basis for subsequent prediction and LOD adaptive adjustment; The intelligent obstacle detection and prediction module uses quantified key indicators as input features and feeds them into a pre-trained machine learning model. The model then intelligently analyzes the dynamic changes in the scene ahead of the trainee and accurately assesses whether there are obstacles approaching at high speed. The LOD intelligent adaptive adjustment module automatically prioritizes increasing the level of detail in the obstacle area when it detects an obstacle is about to appear quickly in front of the trainee. In the initial stage of the obstacle approaching at high speed, low-precision LOD combined with motion blur is used. As the obstacle approaches the train further, it gradually and smoothly transitions to high-precision LOD to ensure the continuity of the visual effect.

Images