A safety control method, system, electronic equipment, media, and program product based on a sightseeing miniature train.

Abstract

This invention belongs to the field of data processing technology and discloses a safety control method, system, electronic device, medium, and program product based on a sightseeing train. The method includes: firstly, dividing the operating route into segments and collecting information on fixed obstructions in each segment to generate an occlusion risk map containing primary attributes such as obstruction type, obstruction viewpoint, and historical risk level; then acquiring the onboard visual image, operating status parameters, time, date, and weather of the current segment of the train; using a pre-trained CNN-LSTM hybrid model, with the CNN branch processing the primary and secondary attributes and the LSTM branch processing the secondary and tertiary attributes, comprehensively scoring the occlusion risk of the current segment; if the risk reaches a medium to high level, outputting warning commands such as voice prompts, speed adjustments, or risk markers, which can improve the safety of the sightseeing train.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a safety control method, system, electronic device, medium, and program product based on a sightseeing miniature train. Background Technology

[0002] The sightseeing train travels through open spaces such as parks and woodlands. During its journey, it is often obstructed by sunshades, landscape decorations, billboards, etc. Pedestrians suddenly appearing from the obstructed areas can easily cause collisions. Traditional solutions rely on pedestrian detection and brake in time when a pedestrian suddenly appears in front of the train. However, they cannot make predictive judgments based on environmental conditions. In addition, traditional solutions mainly rely on real-time image data for detection and early warning, which requires high computing resources and is not timely enough in predictive response. Summary of the Invention

[0003] The purpose of this invention is to address the above-mentioned problems by providing a safety control method, system, electronic device, medium, and program product based on a sightseeing train, which at least partially solves the above-mentioned problems.

[0004] According to a first aspect of the present invention, a safety control method based on a sightseeing train is provided, comprising: segmenting the route of the sightseeing train; collecting information on fixed obstructions in each segment; extracting a first attribute of the obstruction, wherein the first attribute includes at least the type of obstruction, the obstruction angle, and the historical risk level; acquiring a second, third, and fourth attribute of the train in the current segment, wherein the second attribute includes an onboard visual image of the train, the third attribute includes train operating status parameters, and the fourth attribute includes the current time, date, and weather conditions; comprehensively scoring the obstruction risk of the current segment based on the first, second, third, and fourth attributes using a pre-trained machine learning model, wherein the machine learning model is a hybrid model of CNN-LSTM structure, wherein the CNN branch processes the first and second attributes, and the LSTM branch processes the third and fourth attributes; matching the scoring result with a preset risk level threshold, and if the risk level reaches a medium-high level, outputting a corresponding warning instruction, wherein the warning instruction includes a voice prompt, speed adjustment, and / or risk marking prompt.

[0005] According to a second aspect of the present invention, a safety control system based on a sightseeing train is provided, comprising: a first processing module, configured to perform path segmentation processing on the operating route of the sightseeing train, collect information on fixed obstructions in each section, and extract a first attribute of the obstruction, wherein the first attribute includes at least the type of obstruction, the obstruction angle, and the historical risk level; a first acquisition module, configured to acquire a second attribute, a third attribute, and a fourth attribute of the train in the current section, wherein the second attribute includes an onboard visual image of the train, the third attribute includes train operating status parameters, and the fourth attribute includes the current time, date, and weather conditions; a prediction module, configured to comprehensively score the obstruction risk of the current section based on the first attribute, the second attribute, the third attribute, and the fourth attribute using a pre-trained machine learning model, wherein the machine learning model is a hybrid model of CNN-LSTM structure, wherein the CNN branch processes the first attribute and the second attribute, and the LSTM branch processes the third attribute and the fourth attribute; and an early warning module, configured to match the scoring result with a preset risk level threshold, and if the risk level reaches a medium-high level, output a corresponding early warning instruction, wherein the early warning instruction includes a voice prompt, a speed adjustment, and / or a risk marker prompt.

[0006] According to a third aspect of the invention, a computing device is provided, the computing device comprising: at least one processing unit; at least one memory coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions, when executed by the at least one processing unit, enabling the computing device to perform steps according to the method of the first aspect.

[0007] According to a fourth aspect of the present invention, a computer-readable storage medium is also provided. The computer-readable storage medium stores a computer program that, when executed by a machine, performs the method of the first aspect of the present invention.

[0008] According to a fifth aspect of the present invention, a computer program product is also provided, comprising a computer program that, when executed by a machine, performs the method of the first aspect of the present invention.

[0009] Therefore, this invention proposes a safety control method and system based on a sightseeing train. This method generates an occlusion risk map by segmenting the path and collecting information on fixed obstructions along each segment. It then combines onboard visual images, operational status parameters, and temporal environmental parameters, using a pre-trained CNN-LSTM hybrid model to comprehensively score the occlusion risk of the current segment. This method not only considers the impact of static obstructions within the scenic area but also integrates real-time dynamic factors (such as temporary obstructions and weather changes) and vehicle operating status, achieving a comprehensive assessment and prediction of occlusion risks. Compared to traditional solutions that rely solely on real-time image data to monitor pedestrians, this invention can provide early warnings and take preventative measures, effectively reducing the risk of collisions caused by sudden pedestrian appearances, while also reducing computational demands and improving response timeliness.

[0010] Furthermore, by weighted fusion of spatial and temporal features, this method can adaptively focus on key risk factors, improving the accuracy and timeliness of early warnings. Based on the risk scoring results, different levels of warning instructions can be output, including voice prompts, speed adjustments, or risk marker prompts, thereby realizing a tiered risk intervention strategy. Overall, this invention improves the safety of sightseeing train operation and reduces safety hazards caused by visual obstruction.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0012] Figure 1 A flowchart illustrating a safety control method based on a sightseeing train according to an embodiment of the present invention is shown.

[0013] Figure 2 A schematic diagram of the CNN-LSTM model structure according to an embodiment of the present invention is shown.

[0014] Figure 3 A schematic diagram of a safety control system based on a sightseeing train according to an embodiment of the present invention is shown;

[0015] Figure 4 A schematic diagram of an electronic device structure according to an embodiment of the present invention is shown. Detailed Implementation

[0016] The present invention will now be described in detail with reference to the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, methodological, or functional modifications made by those skilled in the art based on these embodiments are included within the scope of protection of the present invention.

[0017] The scope of the embodiments described herein includes the entire scope of the claims and all available equivalents thereof. Throughout this document, the terms “first,” “second,” etc., are used only to distinguish one element from another, without requiring or implying any actual relationship or order between the elements. The terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion and do not exclude the presence of additional identical elements in the structure, apparatus, or device that includes the stated element. The various embodiments described herein are presented in a progressive manner, with each embodiment focusing on its differences from other embodiments; similar or identical parts between embodiments can be referred to interchangeably.

[0018] Figure 1 A flowchart illustrating a safety control method based on a sightseeing train according to an embodiment of the present invention is shown. It should be understood that the method may also include additional steps not shown and / or the steps shown may be omitted, and the scope of the invention is not limited in this respect.

[0019] In step 101, the route of the sightseeing train is segmented, information on fixed obstructions in each segment is collected, and the first attribute of the obstruction is extracted. The first attribute includes at least the type of obstruction, the obstruction angle, and the historical risk level.

[0020] In some embodiments, the sightseeing train route is divided into several continuous segments (such as "entrance plaza segment", "flower sea boardwalk segment", "folk custom street segment") according to the functional zoning of the scenic area, road characteristics (such as straight / curved roads, open / narrow sections) or the density of obstructions. Each segment is assigned a unique identifier (such as an ID number) and basic information such as the coordinates of the start and end points, length, and speed limit is recorded.

[0021] Because the sightseeing train route is relatively stable, the obstacles along the route, such as permanent landscape facilities (sculptures, greenery, signs), visitor rest pavilions, and flower beds, are long-term fixed objects whose attributes (such as height, width, location, and obstruction angle) will not change significantly for months or even years. If these static attributes are obtained in real time, the system will repeatedly process redundant information, wasting computing power.

[0022] By generating occlusion risk maps in advance, this fixed information can be collected and stored all at once, serving as benchmark data for risk assessment and significantly reducing the burden of real-time calculations.

[0023] For each segment, data on all long-standing fixed obstructions are collected through on-site surveys, laser scanning, or historical image analysis. The primary attribute is extracted, including, for example, obstruction type: categorized by physical form (e.g., sculptures, green walls, signs, visitor rest pavilions, etc.), recorded using a coding system (e.g., 1 = sculpture, 2 = green wall); obstruction angle: calculated by simulating the perspective of a moving train, determining the obstruction angle of the obstruction to the onboard camera's field of view (e.g., a 3-meter-wide billboard at a distance of 5 meters has an obstruction angle of 60°), quantified as a value from 0 to 180°; and historical risk level: based on past incidents of pedestrian crossing accidents or visitor complaints around the obstruction, the risk level is divided into 1-5 levels (level 1 being the lowest, level 5 the highest), serving as a historical reference for risk assessment.

[0024] By integrating the locations of fixed obstructions (latitude and longitude or distance relative to the starting point of the road segment), primary attributes, and basic road segment information for each road segment, a digital obstruction risk map is constructed. The map is stored in vector layer format and can be matched in real time with the current road segment and corresponding fixed obstruction information through the positioning system of the miniature train (such as GPS).

[0025] In step 102, the second, third, and fourth attributes of the train on the current route are obtained. The second attribute includes the train's onboard visual image, the third attribute includes the train's operating status parameters, and the fourth attribute includes the current time, date, and weather conditions.

[0026] While the primary attribute of fixed obstructions is marked in advance on the map, the actual risk can change due to real-time factors. For example, the map may indicate that vegetation on a certain road section obstructs the view by 30°, but heavy rain can reduce visibility, potentially increasing the obstruction to 45° and further amplifying the danger of blind spots. The map only includes fixed obstructions, but scenic areas may temporarily erect advertising arches, booths, etc. (which are temporary obstructions) that are not pre-stored on the map. The ultimate severity of the obstruction risk depends not only on the obstruction itself but also on the real-time operating status of vehicles. For instance, during holidays or peak hours, the high density of people increases the probability of pedestrians crossing obstructions, thus raising the risk index.

[0027] In some embodiments, regarding vehicle-mounted visual images, for example, high-definition cameras (frame rate 10-20 frames / second) installed on the front and sides of the train body capture real-time images of the current road segment. After preprocessing (e.g., defogging, noise reduction), key dynamic information is extracted: the location, size, and time of appearance of temporary obstructions (e.g., temporary lantern frames or mobile vending stalls during festivals); and dynamic cues around the obstructions (e.g., pedestrian limbs, moving shadows, and object swaying behind the obstructions).

[0028] Regarding operating status parameters, for example, the dynamic data of the train is acquired in real time through onboard sensors, including the current speed (km / h), acceleration (m / s²), distance from fixed obstructions in front, and current lane position (e.g., driving on the left / right), reflecting the real-time motion status of the vehicle.

[0029] Regarding time and environmental parameters, for example, the current time / date is used to associate tourist flow patterns and tourist density information at different times and on different road sections in the past (e.g., peak hours are from 10:00 to 16:00 on weekends); weather conditions include weather types such as sunny / rainy / foggy / cloudy, light intensity (e.g., night / cloudy), rainfall, etc., which are quantified as parameters that affect the accuracy of visual recognition (e.g., low light intensity on rainy days reduces visual clarity).

[0030] In step 103, based on the first attribute, the second attribute, the third attribute, and the fourth attribute, a comprehensive score is given for the occlusion risk of the current road segment using a pre-trained machine learning model. The machine learning model is a hybrid CNN-LSTM model, wherein the CNN branch processes the first and second attributes, and the LSTM branch processes the third and fourth attributes.

[0031] In some embodiments, a pre-trained machine learning model is used to integrate static data of the scenic area with dynamic real-time information of the miniature train and time environment information to output a quantitative risk score.

[0032] Specifically, in some embodiments, the machine learning model employs a CNN-LSTM hybrid structure, such as... Figure 2 As shown, it includes an input layer, a CNN branch, an LSTM branch, a feature fusion layer, and an output layer.

[0033] The CNN-LSTM model is a deep learning architecture that combines a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network. This combination is particularly suitable for processing data that has both spatial structural information (such as local features in an image, or other types of features that are related to each other) and temporal dependencies (such as time series data).

[0034] CNNs are primarily used to extract spatial features from input data. They identify patterns or features in input data (such as images) through a series of convolutional layers. A typical CNN includes convolutional layers, pooling layers, and fully connected layers. Convolutional layers use filters to scan the entire input data to detect specific types of features; pooling layers are typically used to reduce the dimensionality of the data while retaining the most important feature information; and fully connected layers are responsible for mapping the features learned from previous layers to the output.

[0035] LSTM is a special type of Recurrent Neural Network (RNN) specifically designed to learn dependencies over long time intervals. It effectively captures long-term dependencies in time-series data, overcoming the vanishing gradient problem encountered by traditional RNNs when processing long sequences. An LSTM unit contains three gating mechanisms: a forget gate, an input gate, and an output gate. These gating mechanisms help the network decide which information should be remembered or forgotten, thus effectively managing the flow of information within the memory unit.

[0036] For example, based on the above model structure, the specific prediction process includes:

[0037] Step 1031: Input layer preprocesses input data.

[0038] For the first attribute: Extract fixed occlusion information for the current road segment from the occlusion risk map and convert it into a spatial feature map:

[0039] Type of obstruction (sculpture / greenery / signage, etc., using one-hot encoding, e.g., sculpture=[1,0,0]);

[0040] Occlusion angle (0-180°, normalized to the 0-1 range, e.g., 60°→0.33);

[0041] Historical risk levels (levels 1-5, normalized to the 0-1 range, e.g., level 4 → 0.8).

[0042] For the second attribute: After preprocessing, the location, size and dynamic cue features around the temporary occlusion in the vehicle visual image are extracted to form a dynamic feature map;

[0043] For the third attribute, the vehicle's operating parameters (speed, acceleration, and distance to obstacles in front) are normalized to form a 3-dimensional vector.

[0044] For the fourth attribute, the time environment parameters: current time (marked as 1 during peak hours and 0 during off-peak hours), date (marked as 1 during holidays and 0 during non-holiday hours), and weather conditions (one-hot encoded) are converted into multi-dimensional time series vectors and composed of time series sequences according to time steps.

[0045] Optionally, the spatial feature map and the dynamic feature map can be processed into a unified mesh feature map, specifically including,

[0046] Step 201, Physical Space Meshization

[0047] Centered on the current position of the toy train, a predefined area (covering the possible path the train may travel and the obstructions on both sides) is defined as the physical spatial range of the feature map.

[0048] The area is divided into A two-dimensional grid (e.g., 50 meters horizontally ÷ 0.1 meters = 500 grids, 10 meters vertically ÷ 0.1 meters = 100 grids), where each grid represents a 0.1 meter × 0.1 meter area in reality (the precision can be adjusted according to the specific scenario; this is just an example).

[0049] Establish the correspondence between grid indexes and real-world coordinates (e.g., grid (200, 50) corresponds to the real-world position of "20 meters ahead, 0 meters to the left").

[0050] Step 202: Fill in the first attribute to obtain the dynamic feature map.

[0051] Each grid is filled with several fixed attributes such as occlusion type, occlusion viewpoint, and historical risk level.

[0052] Step 203: Fill in part of the second attribute to obtain the spatial feature map.

[0053] Each grid is filled with dynamic attributes such as the location and size of the temporary occlusion and the dynamic cue features around the occlusion.

[0054] The fixed and dynamic attributes of each grid ultimately form a three-dimensional tensor of width × height × number of channels, i.e., a unified grid feature map. Here, "width × height" corresponds to the grid distribution in physical space, and "channels" correspond to different attribute dimensions.

[0055] Grid feature maps pixelate the physical space through grid division, with each grid corresponding to a small region in reality. The attributes within that region are then filled into the grid, ultimately forming a panoramic image of spatial attributes. This allows CNNs to automatically capture key spatial feature correlations through convolutional operations. Furthermore, the parallel computation efficiency of convolutional operations on grid feature maps is higher, further improving the timeliness of risk assessment.

[0056] Step 1032, CNN branch processing

[0057] The CNN branch processes spatial and dynamic feature maps, outputting spatial feature vectors that capture the spatial distribution and relationships of occluded objects. Specifically, as an example, the CNN branch consists of two convolutional blocks (each containing a convolutional layer, a batch normalization layer, and a max-pooling layer) and a fully connected layer. The convolutional layers perform convolution operations on the input features using 3×3 convolutional kernels to extract spatial correlation features. After compression by the pooling layer, the fully connected layer outputs the spatial feature vector.

[0058] Step 1033, LSTM branch processing

[0059] The LSTM branch processes the vehicle operating parameter vector and the time-series sequence of environmental parameters, outputting a time-series feature vector to capture the dynamic changes of risk factors. Specifically, as an example, the LSTM branch consists of one bidirectional LSTM layer and a fully connected layer. The LSTM layer processes the input sequence, capturing the dynamic changes of vehicle operating parameters and environmental parameters, and outputting a time-series feature vector.

[0060] Step 1034, Feature Fusion

[0061] The feature fusion layer concatenates spatial feature vectors with temporal feature vectors and obtains a fused feature vector through weighted fusion.

[0062] Optionally, the feature fusion layer concatenates the spatial feature vector with the temporal feature vector, and obtains the fused feature vector by adjusting the weights through an attention mechanism.

[0063] The attention mechanism dynamically calculates weights, enabling the model to adaptively focus on key features and ignore secondary features. The attention mechanism can be represented as:

[0064]

[0065] Where Q (query): the feature currently being focused on; K (key): all possible features; V (value): the actual feature value corresponding to the key; scaling factor: Used to stabilize gradients and prevent the softmax gradient from vanishing due to an excessively large inner product result.

[0066] Specifically, the implementation of the attention mechanism includes,

[0067] (1) Feature splicing and projection

[0068] The spatial feature vector output by the CNN ( ) and the time-series feature vector output by LSTM ( splicing:

[0069]

[0070] By projecting it into a low-dimensional space using a fully connected layer, computational complexity is reduced.

[0071]

[0072] in, It is a projection matrix. It is a bias term.

[0073] (2) Calculation of attention weights

[0074] The projected features are mapped to scalar values ​​(i.e., attention scores) through another fully connected layer:

[0075]

[0076] in, , It is a weight matrix. , It is a bias term. As an activation function, it ensures that the score is in the range [-1, 1].

[0077] Applying the softmax function to the attention scores at all time steps normalizes them to the weight distribution:

[0078]

[0079] in, This represents the importance weight of the i-th feature.

[0080] (3) Weighted feature fusion

[0081] Normalized weights Apply to original splicing features The weighted fusion feature vector is obtained as follows:

[0082]

[0083] This vector, through weighted aggregation, highlights the impact of key risk features while suppressing irrelevant information.

[0084] 5. Output layer

[0085] The fused feature vectors are processed by the output layer and mapped to a comprehensive score of 0-100. The higher the score, the higher the occlusion risk of the current road segment.

[0086] For example, the output layer adopts a two-level fully connected architecture, progressively compressing the feature dimensions and mapping them to the risk scoring space, specifically including,

[0087] Feature dimensionality reduction: The fused feature vector output by the attention mechanism is compressed through a fully connected layer to reduce the risk of overfitting and extract higher-order feature relationships.

[0088]

[0089] in, It is a weight matrix. It is a bias term; ReLU introduces nonlinearity to the activation function.

[0090] Risk score mapping: The above features are further compressed to 1 dimension, and the output value is constrained to the [0,1] interval by using the Sigmoid activation function:

[0091]

[0092] in, It is a weight matrix. It is a bias term.

[0093] Scoring calibration: Linearly map normalized scores to the 0-100 score range.

[0094] Step 104: Match the scoring results with the preset risk level threshold. If the risk level reaches the medium to high level, output the corresponding warning instructions, including voice prompts, running speed adjustment or risk mark prompts.

[0095] In some embodiments, the comprehensive score of 0-100 is divided into three risk thresholds, for example:

[0096] Low risk: Score 0-30, no active intervention required;

[0097] Medium risk: Score 31-60, basic early warning required;

[0098] High risk: Score 61-100, requires strong intervention.

[0099] The output layer generates a comprehensive score of 0-100, which is compared with a preset three-level threshold to determine the current risk level. Based on the severity of the risk level, tiered warning instructions are triggered, for example:

[0100] For medium-risk levels (31-60 points), the approach is primarily "suggestive intervention," using information dissemination to reduce risk.

[0101] Voice prompts: Issue concise prompts to the driver (e.g., "Medium-risk area, drive with caution");

[0102] Risk marker alert: The location of currently high-risk obstructions is marked with a yellow icon on the electronic map displayed on the vehicle, helping the driver to pay close attention.

[0103] High-risk level (61-100 points) instruction output

[0104] Primarily through "mandatory intervention," risks are reduced through proactive control:

[0105] Enhance voice prompts: Provide clear instructions to the driver (e.g., "High risk! We recommend reducing speed to 5 km / h");

[0106] Operating speed adjustment: Automatically sends speed adjustment commands to the power module to extend braking distance and reserve reaction buffer time;

[0107] Enhanced risk marking: The display screen uses flashing red icons to indicate the location of obstructions.

[0108] Based on the characteristics of data with different attributes, the embodiments of this application use CNN and LSTM to process different attributes respectively. CNN uses its spatial extraction capability to process static / spatial data, and LSTM uses its temporal modeling capability to process dynamic / temporal data. Then, the features of the two are associated through a fusion layer. This structure is more in line with the nature of multimodal data, can reduce information loss, reduce the difficulty of model learning, and ultimately improve the prediction accuracy of complex scenes (such as occlusion risk assessment).

[0109] like Figure 3 As shown, this embodiment of the invention provides a safety control system 300 based on a sightseeing train, including:

[0110] The first processing module 301 is used to perform path segmentation processing on the operation route of the sightseeing train, collect information on fixed obstructions in each section, and extract the first attribute of the obstruction. The first attribute includes at least the type of obstruction, the obstruction angle, and the historical risk level.

[0111] The first acquisition module 302 is used to acquire the second attribute, the third attribute and the fourth attribute of the small train in the current section. The second attribute includes the on-board visual image of the small train, the third attribute includes the running status parameters of the small train, and the fourth attribute includes the current time, date and weather conditions.

[0112] The prediction module 303 is used to comprehensively score the occlusion risk of the current road segment based on the first attribute, the second attribute, the third attribute and the fourth attribute, through a pre-trained machine learning model. The machine learning model is a hybrid model of CNN-LSTM structure, wherein the CNN branch processes the first attribute and the second attribute, and the LSTM branch processes the third attribute and the fourth attribute.

[0113] The early warning module 304 is used to match the scoring results with a preset risk level threshold. If the risk level reaches the medium-high level, the module outputs a corresponding early warning instruction, which includes voice prompts, running speed adjustment and / or risk marking prompts.

[0114] like Figure 4 As shown, an embodiment of the present invention provides an electronic device, including a processor and a memory, wherein the memory stores program instructions, and the processor executes the program instructions to implement the safety control method based on a sightseeing train as described in any of the preceding embodiments.

[0115] Another embodiment of this application provides a computer-readable storage medium having computer program instructions stored thereon, which can be executed by a processor to implement the methods and / or technical solutions of any one or more embodiments of this application described above.

[0116] In particular, the methods and / or embodiments in this application can be implemented as computer software programs. For example, the embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowchart. When the computer program is executed by a processor, it performs the functions defined in the methods of this application.

[0117] It should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0118] The detailed descriptions listed above are merely specific descriptions of feasible embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. All equivalent embodiments or modifications made without departing from the spirit of the present invention should be included within the scope of protection of the present invention.

Claims

1. A safety control method based on a sightseeing miniature train, characterized in that, include: The sightseeing train route is segmented according to the scenic area's functional zones, road characteristics, and the density of obstructions. Information on fixed obstructions in each section is collected, and the first attribute of the obstruction is extracted. The first attribute includes at least the type of obstruction, the obstruction angle, and the historical risk level. By integrating the locations of fixed obstructions, primary attributes, and basic information of each road segment, a digital obstruction risk map is constructed. The second, third, and fourth attributes of the train on the current route are obtained. The second attribute includes the train's onboard visual image, the third attribute includes the train's operating status parameters, and the fourth attribute includes the current time, date, and weather conditions. After preprocessing, the onboard visual image of the train is used to extract the location, size, and dynamic cue features around the temporary obstructions to form a dynamic feature map. The first attribute of the current road segment is extracted from the occlusion risk map. The first and second attributes are processed into a unified grid feature map. Specifically, with the current position of the train as the center, a preset area is delineated to cover the possible path of the vehicle and the obstructions on both sides. This area is divided into an M×N two-dimensional grid. The correspondence between the two-dimensional grid and the real coordinates is established. The first attribute feature and the second attribute feature are filled into each grid respectively to form a three-dimensional tensor of width × height × number of channels as the grid feature map. The third and fourth attributes are combined into a time-series vector sequence according to time steps. Based on the first attribute, the second attribute, the third attribute, and the fourth attribute, a pre-trained machine learning model is used to comprehensively score the occlusion risk of the current road segment. The machine learning model is a hybrid model with a CNN-LSTM structure, wherein the CNN branch processes the first and second attributes, and the LSTM branch processes the third and fourth attributes. The scoring results are matched with preset risk level thresholds. If the risk level reaches the medium to high level, a corresponding warning instruction is output. The warning instruction includes voice prompts, running speed adjustment, and / or risk marker prompts.

2. The safety control method based on a sightseeing train according to claim 1, characterized in that, include: The hybrid CNN-LSTM model includes a CNN branch, an LSTM branch, a feature fusion layer, and an output layer. The CNN branch consists of two convolutional blocks and a fully connected layer. Each convolutional block contains a convolutional layer, a batch normalization layer, and a max-pooling layer, processing the first and second attribute features to output a spatial feature vector. The LSTM branch contains a bidirectional LSTM layer and a fully connected layer, processing the third and fourth attribute features to output a temporal feature vector. The feature fusion layer concatenates the spatial and temporal feature vectors and obtains a fused feature vector through weighted fusion. The output layer processes the fused feature vector, mapping it to a comprehensive score of 0-100, where a higher score indicates a higher occlusion risk for the current road segment.

3. The safety control method based on a sightseeing train according to claim 2, characterized in that, include: The process of obtaining the fused feature vector through weighted fusion includes: concatenating the spatial feature vector output by the CNN and the temporal feature vector output by the LSTM to obtain the original concatenated feature; projecting it into a low-dimensional space through a fully connected layer; mapping the projected feature into attention scores through another fully connected layer; applying the softmax function to the attention scores at all time steps to normalize them into a weight distribution; and applying the normalized weights to the original concatenated feature to obtain the weighted fused feature vector.

4. The safety control method based on a sightseeing train according to claim 3, characterized in that, include: The output layer adopts a two-level fully connected architecture, which gradually compresses the feature dimension and maps it to the risk scoring space. Specifically, it compresses the fused feature vector output by the attention mechanism through the fully connected layer to obtain low-dimensional features, further compresses the low-dimensional features to 1 dimension, and constrains the output value score to the [0,1] interval through the Sigmoid activation function, and linearly maps the normalized score to the 0-100 score interval.

5. A safety control system based on a sightseeing miniature train, characterized in that, include: The first processing module is used to segment the route of the sightseeing train according to the functional zoning of the scenic area, road characteristics and the distribution density of obstructions, collect information on fixed obstructions in each section, and extract the first attribute of the obstruction. The first attribute includes at least the type of obstruction, the obstruction angle and the historical risk level. By integrating the locations of fixed obstructions, primary attributes, and basic information of each road segment, a digital obstruction risk map is constructed. The first acquisition module is used to acquire the second, third, and fourth attributes of the train on the current route. The second attribute includes the train's onboard visual image, the third attribute includes the train's operating status parameters, and the fourth attribute includes the current time, date, and weather conditions. After preprocessing, the onboard visual image of the train is used to extract the location, size, and dynamic cue features around the temporary obstruction to form a dynamic feature map. The second processing module is used to extract the first attribute of the current road segment from the occlusion risk map, and process the first attribute and the second attribute into a unified grid feature map. Specifically, with the current position of the train as the center, a preset area is delineated to cover the possible path of the vehicle and the obstructions on both sides. This area is divided into an M×N two-dimensional grid. The correspondence between the two-dimensional grid and the real coordinates is established. The first attribute feature and the second attribute feature are filled into each grid respectively to form a three-dimensional tensor of width × height × number of channels as the grid feature map. The third attribute and the fourth attribute are combined into a time-series vector sequence according to time steps. The prediction module is used to comprehensively score the occlusion risk of the current road segment based on the first attribute, the second attribute, the third attribute and the fourth attribute, through a pre-trained machine learning model. The machine learning model is a hybrid model with a CNN-LSTM structure, wherein the CNN branch processes the first attribute and the second attribute, and the LSTM branch processes the third attribute and the fourth attribute. The early warning module is used to match the scoring results with preset risk level thresholds. If the risk level reaches the medium to high level, the module outputs a corresponding early warning instruction, which includes voice prompts, running speed adjustment, and / or risk marker prompts.

6. An electronic device, the electronic device comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, characterized in that: the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1-4.

7. A computer-readable medium having computer program instructions stored thereon, characterized in that: The computer program instructions can be executed by a processor to implement the method as described in any one of claims 1-4.

8. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, it implements the method as described in any one of claims 1-4.

Images