A safety protection system and method for urban railway platform doors based on laser detection

Abstract

The present application discloses a system and method for protecting the platform door of a city railroad based on laser detection, and relates to the field of laser radar detection technology. The method comprises: after collecting a three-dimensional scanning model of an obstacle through a laser radar, the three-dimensional scanning model is analyzed at the back end using a data processing algorithm based on artificial intelligence and deep learning, so as to capture the outline and essential joint feature representation of the obstacle, and use this feature representation to identify the type of obstacle, so as to make a corresponding alarm based on the identification result, and can use artificial intelligence and deep learning technology to more accurately identify and detect obstacles between the platform door and the train, and distinguish different types of obstacles, thereby reducing the possibility of false alarms and missed alarms, and realizing more intelligent safety protection for the platform door of the city railroad.

Description

Technical Field

[0001] The present application relates to the field of laser radar detection technology, and in particular, to a system and method for protecting platform doors of urban railways based on laser detection. Background Art

[0002] 3D laser radar detection technology is a radar system that emits laser beams to detect the three-dimensional spherical coordinates of the target. It uses the principles of coherent angle measurement and frequency modulation ranging to transmit detection signals to the target, and then compares the received echo signal reflected from the target with the transmitted signal. After corresponding processing, the spherical coordinates of the target can be obtained. 3D laser radar detection technology is currently widely used in many fields such as unmanned driving, space measurement, and topographic mapping. In the railway field, 3D laser detection radar technology can be used to detect obstacles in the blind spot of platform doors to prevent people or other foreign objects from being caught when the safety door is closed, so as to improve the safety performance of the safety door.

[0003] For example, the Chinese patent application number 201810235642.8 discloses a system and method for preventing platform doors and trains from being pinched. When both the platform doors and train doors are closed, the system uses a three-dimensional laser radar to perform a three-dimensional scan of obstacles in the area between the platform doors and the train, detects whether there are obstacles, and classifies and alarms the types of obstacles. Compared with traditional detection methods, this method can detect obstacles and alarm in the blind area between the platform doors and the train, improving the safety of subway operation.

[0004] However, the above-mentioned anti-pinch system for platform doors and trains mainly focuses on the detection of obstacles, and does not explore in depth how to efficiently identify and classify obstacles. Specifically, the above-mentioned anti-pinch system classifies obstacles based on the shape information and displacement information of the obstacles to determine whether the obstacles are living things. Although three-dimensional lidar can provide detailed three-dimensional data, there are still some limitations in practical applications. For example, the existing system may not be able to accurately distinguish different types of obstacles, such as the difference between passengers and different types of leftover items, which may lead to false alarms or missed alarms. In addition, in complex environmental conditions or when facing multiple obstacles, obstacle detection and type identification through lidar information may be interfered with, affecting the detection effect.

[0005] Therefore, an optimized urban railway platform door safety protection system is needed to solve the above problems. Summary of the invention

[0006] This summary is provided to introduce concepts in a brief form that will be described in detail in the detailed description below. This summary is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0007] In a first aspect, the present application provides a safety protection system for urban rail platform doors based on laser detection, the system comprising:

[0008] The obstacle 3D scanning model acquisition module is used to obtain the 3D scanning model of the obstacle collected by the laser radar;

[0009] An obstacle feature extraction module, used for inputting the three-dimensional scanning model of the obstacle into a double-branch boundary information loss compensation network including a boundary feature extraction branch and a trunk network to obtain an obstacle contour feature map and an obstacle essential feature map;

[0010] A feature gradient correlation feature saliency enhancement module, used for inputting the obstacle contour feature map into a mask saliency device based on feature gradient correlation distribution to obtain an obstacle contour salient feature map;

[0011] An obstacle multi-scale feature joint perception module, used for performing feature joint significant semantic perception on the obstacle contour significant feature map and the obstacle essential feature map to obtain an obstacle multi-scale significant joint perception feature map;

[0012] The obstacle detection alarm module is used to determine the type label of the obstacle based on the multi-scale significant joint perception feature map of the obstacle, and determine whether to generate an alarm signal.

[0013] Optionally, the feature gradient associated feature significance enhancement module includes: a multi-directional gradient value distribution calculation unit, which is used to calculate the multi-directional gradient value distribution of each position in the obstacle contour feature map, and determine the gradient amplitude value of each position in the obstacle contour feature map based on the multi-directional gradient value distribution of each position to obtain an obstacle contour feature gradient amplitude distribution feature map; a gradient amplitude local description operator calculation unit, which is used to calculate the gradient amplitude local description operator of each position in the obstacle contour feature gradient amplitude distribution feature map to obtain an obstacle contour feature gradient amplitude local significant distribution feature map; a gated mask calculation unit, which is used to input the obstacle contour feature gradient amplitude local significant distribution feature map into a gated masker based on a GELU function to obtain an obstacle contour feature gradient amplitude local significant gated mask feature map; a feature map position point multiplication calculation unit, which is used to calculate the position point multiplication between the obstacle contour feature gradient amplitude local significant gated mask feature map and the obstacle contour feature map to obtain the obstacle contour significant feature map.

[0014] Optionally, the multi-directional gradient value distribution calculation unit is used to: calculate the gradient amplitude value along the height direction for each position feature value in the obstacle contour feature map to obtain an obstacle contour feature height gradient distribution feature map; calculate the gradient amplitude value along the width direction for each position feature value in the obstacle contour feature map to obtain an obstacle contour feature width gradient distribution feature map; calculate the gradient amplitude value along the channel direction for each position feature value in the obstacle contour feature map to obtain an obstacle contour feature channel gradient distribution feature map; calculate the binary norm between the corresponding position feature values ​​in the obstacle contour feature height gradient distribution feature map, the obstacle contour feature width gradient distribution feature map and the obstacle contour feature channel gradient distribution feature map to obtain the obstacle contour feature gradient amplitude distribution feature map.

[0015] Optionally, the gradient amplitude local description operator calculation unit is used to: determine the local neighborhood range of the gradient amplitude, and calculate a sequence of the gradient amplitude distribution neighborhood feature map of the obstacle contour feature based on the local neighborhood range of the gradient amplitude and the gradient amplitude distribution feature map of the obstacle contour feature; count the number of feature maps in the sequence of the gradient amplitude distribution neighborhood feature map of the obstacle contour feature to obtain the number statistics of the gradient amplitude distribution neighborhood of the obstacle contour feature; calculate the number of feature maps in the sequence of the gradient amplitude distribution feature map of the obstacle contour feature and the gradient amplitude distribution neighborhood feature map of the obstacle contour feature The position differences between the obstacle contour feature gradient amplitude distribution neighborhood feature maps are divided by the obstacle contour feature gradient amplitude distribution feature map by position point to obtain a sequence of obstacle contour feature gradient relative amplitude distribution neighborhood feature maps; the position point addition of each obstacle contour feature gradient relative amplitude distribution neighborhood feature map in the sequence of obstacle contour feature gradient relative amplitude distribution neighborhood feature maps is performed, and the resultant addition is divided by the obstacle contour feature gradient amplitude distribution neighborhood number statistics by position point to obtain the obstacle contour feature gradient amplitude local significant distribution feature map.

[0016] Optionally, the obstacle multi-scale feature joint perception module is used to: input the obstacle contour salient feature map and the obstacle essential feature map into an attention-driven feature joint saliency perception network to obtain the obstacle multi-scale salient joint perception feature map.

[0017] Optionally, the obstacle multi-scale feature joint perception module includes: a first feature shape reshaping unit, used to perform feature shape reshaping on the obstacle contour salient feature map and the obstacle essential feature map to obtain an obstacle contour salient feature shape reshaping matrix and an obstacle essential feature shape reshaping matrix; a feature channel-by-channel interactive perception unit, used to input the obstacle contour salient feature shape reshaping matrix and the obstacle essential feature shape reshaping matrix into the feature channel-by-channel interactive perception module to obtain an obstacle contour-essential feature dependency matrix and an obstacle essence-contour feature dependency matrix; a random deactivation unit, used to input the obstacle contour-essential feature dependency matrix and the obstacle essence-contour feature dependency matrix into the random deactivation module to obtain a pruned optimized obstacle contour-essential feature dependency matrix and a pruned optimized obstacle essence-contour feature dependency matrix; a first matrix multiplication calculation unit, used to input the obstacle contour-essential feature dependency matrix and the obstacle essence-contour feature dependency matrix into the random deactivation module to obtain a pruned optimized obstacle contour-essential feature dependency matrix and a pruned optimized obstacle essence-contour feature dependency matrix; The pruned optimized obstacle contour-essential feature dependency matrix is ​​matrix-multiplied with the obstacle essential feature shape reshaping matrix to obtain a dependency optimized obstacle essential-contour feature dependency matrix; a second matrix multiplication calculation unit is used to multiply the pruned optimized obstacle essential-contour feature dependency matrix with the obstacle contour significant feature shape reshaping matrix to obtain a dependency optimized obstacle contour-essential feature dependency matrix; a second feature shape reshaping unit is used to perform feature shape reshaping on the dependency optimized obstacle essential-contour feature dependency matrix and the dependency optimized obstacle contour-essential feature dependency matrix to obtain an optimized obstacle contour significant feature map and an optimized obstacle essential feature map; a weighted sum calculation unit is used to calculate the weighted sum of the optimized obstacle contour significant feature map and the optimized obstacle essential feature map to obtain the obstacle multi-scale significant joint perception feature map.

[0018] Optionally, the feature channel-by-channel interaction perception unit is used to: perform matrix multiplication on the obstacle contour significant feature shape reshaping matrix and the transposed matrix of the obstacle essential feature shape reshaping matrix, and then divide the matrix by the scale of the obstacle contour-essential feature dependency matrix by position to obtain an obstacle contour-essential feature interaction matrix; perform matrix multiplication on the obstacle contour significant feature shape reshaping matrix and the obstacle essential feature shape reshaping matrix, and then divide the matrix by the scale of the obstacle essential feature shape reshaping matrix by position to obtain an obstacle essential-contour feature interaction matrix; use The obstacle contour-essential feature interaction matrix and the obstacle essence-contour feature interaction matrix are respectively normalized to obtain the obstacle contour-essential feature dependency matrix and the obstacle essence-contour feature dependency matrix.

[0019] Optionally, the obstacle detection alarm module is used to: input the obstacle multi-scale significant joint perception feature map into a classifier-based recognition module to obtain a recognition result, wherein the recognition result is used to represent the type label of the obstacle; and determine whether to generate an alarm signal based on the recognition result.

[0020] In a second aspect, the present application provides a method for protecting the safety of a platform door of an urban railway based on laser detection, the method comprising:

[0021] Obtain a three-dimensional scanning model of the obstacle collected by the laser radar;

[0022] Inputting the three-dimensional scanning model of the obstacle into a double-branch boundary information loss compensation network including a boundary feature extraction branch and a backbone network to obtain an obstacle contour feature map and an obstacle essential feature map;

[0023] Inputting the obstacle contour feature map into a mask salient device based on feature gradient correlation distribution to obtain an obstacle contour salient feature map;

[0024] Performing feature joint saliency semantic perception on the obstacle contour saliency feature map and the obstacle essential feature map to obtain an obstacle multi-scale saliency joint perception feature map;

[0025] Based on the obstacle multi-scale significant joint perception feature map, the type label of the obstacle is determined, and it is determined whether to generate an alarm signal.

[0026] Optionally, the obstacle contour feature map is input into a mask salient device based on feature gradient association distribution to obtain an obstacle contour salient feature map, including: calculating the multi-directional gradient value distribution of each position in the obstacle contour feature map, and determining the gradient amplitude value of each position in the obstacle contour feature map based on the multi-directional gradient value distribution of each position to obtain an obstacle contour feature gradient amplitude distribution feature map; calculating the gradient amplitude local description operator of each position in the obstacle contour feature gradient amplitude distribution feature map to obtain an obstacle contour feature gradient amplitude local significant distribution feature map; inputting the obstacle contour feature gradient amplitude local significant distribution feature map into a gated masker based on a GELU function to obtain an obstacle contour feature gradient amplitude local significant gated mask feature map; calculating the position point multiplication between the obstacle contour feature gradient amplitude local significant gated mask feature map and the obstacle contour feature map to obtain the obstacle contour salient feature map.

[0027] With the above technical solution, after the three-dimensional scanning model of the obstacle is collected by the laser radar, the data processing algorithm based on artificial intelligence and deep learning is used at the back end to analyze the three-dimensional scanning model, so as to capture the outline and essential joint feature representation of the obstacle, and use this feature representation to identify the type of obstacle, so as to issue a corresponding alarm based on the identification result. In this way, artificial intelligence and deep learning technology can be used to more accurately identify and detect obstacles between platform doors and trains, and distinguish different types of obstacles, thereby reducing the possibility of false alarms and missed alarms, and realizing more intelligent safety protection of urban rail platform doors.

[0028] Other features and advantages of the present application will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale. In the drawings:

[0030] Figure 1 It is a block diagram of a safety protection system for urban railway platform doors based on laser detection according to an exemplary embodiment.

[0031] Figure 2 It is a flow chart of a method for safety protection of urban railway platform doors based on laser detection according to an exemplary embodiment.

[0032] Figure 3 It is a block diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0033] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as being limited to the embodiments described herein. Instead, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes and are not intended to limit the scope of protection of the present application.

[0034] It should be understood that the various steps described in the method implementation of the present application can be performed in different orders and / or performed in parallel. In addition, the method implementation may include additional steps and / or omit the steps shown. The scope of the present application is not limited in this respect.

[0035] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0036] It should be noted that the concepts such as "first" and "second" mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0037] It should be noted that the modifications of "one" and "plurality" mentioned in the present application are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0038] The names of the messages or information exchanged between multiple devices in the embodiments of the present application are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0039] The specific implementation of the present application is described in detail below with reference to the accompanying drawings.

[0040] Figure 1 FIG. 1 is a block diagram of a safety protection system for urban rail platform doors based on laser detection according to an exemplary embodiment. Figure 1 As shown, the system 100 includes:

[0041] The obstacle 3D scanning model acquisition module 101 is used to acquire the 3D scanning model of the obstacle acquired by the laser radar;

[0042] The obstacle feature extraction module 102 is used to input the three-dimensional scanning model of the obstacle into a double-branch boundary information loss compensation network including a boundary feature extraction branch and a backbone network to obtain an obstacle contour feature map and an obstacle essential feature map;

[0043] A feature gradient correlation feature saliency enhancement module 103 is used to input the obstacle contour feature map into a mask salient device based on feature gradient correlation distribution to obtain an obstacle contour salient feature map;

[0044] The obstacle multi-scale feature joint perception module 104 is used to perform feature joint saliency semantic perception on the obstacle contour saliency feature map and the obstacle essential feature map to obtain an obstacle multi-scale saliency joint perception feature map;

[0045] The obstacle detection alarm module 105 is used to determine the type label of the obstacle based on the obstacle multi-scale significant joint perception feature map, and determine whether to generate an alarm signal.

[0046] In response to the above technical problems, in the technical solution of this application, a safety protection system for urban rail platform doors based on laser detection is proposed, which can collect the three-dimensional scanning model of the obstacle through the laser radar, and then use the data processing algorithm based on artificial intelligence and deep learning to analyze the three-dimensional scanning model at the back end, so as to capture the outline and essential joint feature representation of the obstacle, and use this feature representation to identify the type of obstacle, so as to make corresponding alarms based on the recognition results, and can use artificial intelligence and deep learning technology to more accurately identify and detect obstacles between platform doors and trains, and distinguish different types of obstacles, thereby reducing the possibility of false alarms and missed alarms, and realizing more intelligent safety protection for urban rail platform doors. In addition, this intelligent recognition and alarm method can also maintain a high obstacle type recognition accuracy in complex environments, which helps to improve the reliability and stability of the entire system.

[0047] Specifically, in the technical solution of the present application, first, a three-dimensional scanning model of an obstacle collected by a laser radar is obtained. Then, the three-dimensional scanning model of the obstacle is input into a dual-branch boundary information loss compensation network including a boundary feature extraction branch and a backbone network to obtain an obstacle contour feature map and an obstacle essential feature map. Among them, the dual-branch boundary information loss compensation network including a boundary feature extraction branch and a backbone network includes two branches, one of which is a boundary feature extraction branch, which focuses on extracting the boundary information of the obstacle. Boundary information is very important for identifying obstacles because it can help understand the shape and outline of the object. In the narrow space between the platform door and the train, accurately identifying the boundary of the obstacle is crucial to prevent people or objects from being trapped. Specifically, in the process of deep learning, especially in convolutional neural networks, boundary information may be lost or become blurred. By compensating for this loss through a special network design, such as a dual-branch boundary information loss compensation network, the integrity of the boundary information can be ensured, which is very critical for accurately identifying obstacles. The other is a backbone network branch that can capture the internal and essential features of the obstacle, such as color, material, and structure. These features are very important for identifying what the obstacle is. By considering both the boundary information and internal features of obstacles, the network can more comprehensively understand the characteristics of obstacles and maintain high detection and recognition quality even in complex environments, which is crucial for correctly classifying obstacles. For example, a person and a backpack may have different shapes and internal features, so the dual-branch network is better able to identify them.

[0048] Then, since the obstacle contour feature map contains the contour feature information about the obstacle, including features such as edge, contour and shape, some of which are of great significance for the type recognition and feature expression of the obstacle, while some are irrelevant interference features and redundant information. Therefore, in order to enhance the key information expression ability in the obstacle contour feature map, improve the understanding of the obstacle contour and the obstacle category recognition ability, in the technical solution of the present application, the obstacle contour feature map is further input into the mask salient device based on the feature gradient correlation distribution to obtain the obstacle contour salient feature map. Specifically, the mask salient device based on the feature gradient correlation distribution is used to enhance the feature map expression ability. Its main function is to highlight the important information in the obstacle contour feature map through the feature gradient amplitude measurement and the gated mask mechanism, thereby improving the performance of the model and the expression ability of the obstacle contour feature. Specifically, firstly, by performing multi-directional gradient calculation on the obstacle contour feature map, it is possible to detect which areas in the feature map have the most obvious edge contours and details, that is, which areas contain more obstacle-related information. Next, based on the results of the gradient calculation, analyze which features are most critical for representing specific obstacle objects or obstacle recognition scenarios. In this solution, it is used to analyze which directions of gradient features are more important for representing obstacle contours. Then, through the gating mechanism, certain areas in the feature map are enhanced or suppressed to ensure that key features related to subsequent obstacle type recognition tasks are retained, while non-critical features are weakened. Finally, through the weighted dot multiplication operation, the enhanced obstacle contour features are combined with the original contour representation features to obtain the obstacle contour salient feature map. This can better identify and distinguish different obstacle contours and improve the model's sensitivity and generalization ability to input data.

[0049] In one embodiment of the present application, the feature gradient associated feature significance enhancement module includes: a multi-directional gradient value distribution calculation unit, which is used to calculate the multi-directional gradient value distribution of each position in the obstacle contour feature map, and determine the gradient amplitude value of each position in the obstacle contour feature map based on the multi-directional gradient value distribution of each position to obtain an obstacle contour feature gradient amplitude distribution feature map; a gradient amplitude local description operator calculation unit, which is used to calculate the gradient amplitude local description operator of each position in the obstacle contour feature gradient amplitude distribution feature map to obtain an obstacle contour feature gradient amplitude local significant distribution feature map; a gated mask calculation unit, which is used to input the obstacle contour feature gradient amplitude local significant distribution feature map into a gated masker based on a GELU function to obtain an obstacle contour feature gradient amplitude local significant gated mask feature map; a feature map position point multiplication calculation unit, which is used to calculate the position point multiplication between the obstacle contour feature gradient amplitude local significant gated mask feature map and the obstacle contour feature map to obtain the obstacle contour significant feature map.

[0050] Furthermore, in one embodiment of the present application, the multi-directional gradient value distribution calculation unit is used to: calculate the gradient amplitude value along the height direction for each position feature value in the obstacle contour feature map to obtain an obstacle contour feature height gradient distribution feature map; calculate the gradient amplitude value along the width direction for each position feature value in the obstacle contour feature map to obtain an obstacle contour feature width gradient distribution feature map; calculate the gradient amplitude value along the channel direction for each position feature value in the obstacle contour feature map to obtain an obstacle contour feature channel gradient distribution feature map; calculate the binary norm between the corresponding position feature values ​​in the obstacle contour feature height gradient distribution feature map, the obstacle contour feature width gradient distribution feature map and the obstacle contour feature channel gradient distribution feature map to obtain the obstacle contour feature gradient amplitude distribution feature map.

[0051] Furthermore, in one embodiment of the present application, the gradient amplitude local description operator calculation unit is used to: determine the local neighborhood range of the gradient amplitude, and calculate a sequence of the obstacle contour feature gradient amplitude distribution neighborhood feature maps based on the local neighborhood range of the gradient amplitude and the obstacle contour feature gradient amplitude distribution feature map; count the number of feature maps in the sequence of the obstacle contour feature gradient amplitude distribution neighborhood feature maps to obtain the obstacle contour feature gradient amplitude distribution neighborhood number statistics; calculate the obstacle contour feature gradient amplitude distribution feature map and the obstacle contour feature gradient amplitude distribution neighborhood The position differences between each obstacle contour feature gradient amplitude distribution neighborhood feature map in the sequence of feature maps are divided by the obstacle contour feature gradient amplitude distribution feature map by position point to obtain a sequence of obstacle contour feature gradient relative amplitude distribution neighborhood feature maps; each obstacle contour feature gradient relative amplitude distribution neighborhood feature map in the sequence of obstacle contour feature gradient relative amplitude distribution neighborhood feature maps is added by position point and divided by the obstacle contour feature gradient amplitude distribution neighborhood number statistics by position point to obtain the obstacle contour feature gradient amplitude local significant distribution feature map.

[0052] Specifically, the obstacle contour feature map is input into a mask salient device based on feature gradient correlation distribution and processed using the following gradient salient feature enhancement formula to obtain an obstacle contour salient feature map;

[0053] Among them, the gradient significant feature enhancement formula is:

[0054]

[0055]

[0056]

[0057]

[0058]

[0059]

[0060]

[0061] in, , , , , and They are respectively the first , , , , and The eigenvalues ​​of the positions, is the first obstacle along the height direction in the obstacle profile feature map The gradient amplitude value at each position, is the first pixel along the width direction in the obstacle profile feature map The gradient amplitude value at each position, is the first The gradient amplitude value at each position, is the first The eigenvalues ​​at the positions, is the first The eigenvalues ​​at the positions, The range is Neighborhood of express The number of elements in , The first The eigenvalues ​​at the positions, represents the Gaussian error linear gating function, The obstacle contour feature gradient amplitude local significant gated mask feature map The eigenvalues ​​at the positions, is the local significant gated mask feature map of the gradient amplitude of the obstacle contour feature, is the obstacle contour feature map, It means point multiplication by position. is the obstacle contour salient feature map.

[0062] Furthermore, considering that the obstacle contour salient feature map and the obstacle essential feature map respectively contain feature information of different levels and aspects about the obstacle. The obstacle contour salient feature map emphasizes the boundary feature information such as the edge, shape and contour of the obstacle, while the essential feature map focuses more on the essential features and semantic information that are more critical for identifying the obstacle type, such as the color, material and structure of the obstacle. Therefore, in order to effectively fuse and interact the boundary features and essential features of the obstacle contour, so as to better perceive and understand the type of the obstacle, in the technical solution of the present application, the obstacle contour salient feature map and the obstacle essential feature map are further input into the attention-driven feature joint saliency perception network to obtain the obstacle multi-scale salient joint perception feature map. Specifically, the attention-driven feature joint saliency perception network is a feature fusion technology that enhances the feature interaction between the obstacle contour salient feature map and the obstacle essential feature map through multiple steps, and finally obtains an obstacle multi-scale salient joint perception feature map that combines the advantages of both. Specifically, first, the feature matrix along the channel dimension in the input obstacle contour salient feature map and the obstacle essential feature map is expanded into a feature vector to facilitate the subsequent channel-by-channel interactive perception operation. Then, the dependency relationship between the two groups of feature vectors, the obstacle contour feature and the essential feature, is calculated to obtain the obstacle contour-essential feature and obstacle essential-contour feature dependency matrix. Then, the dependency matrix is ​​sparsely processed by pruning operation to enhance the generalization ability of the model. Then, the pruned dependency matrix is ​​used to weight the original features to strengthen the features of the most relevant parts of the obstacle perception and recognition tasks, while suppressing unimportant features. Finally, the optimized feature matrix is ​​reshaped into a feature map form, and the weighted sum is calculated to obtain the obstacle multi-scale salient joint perception feature map. In this way, the intuitive contour boundary information in the obstacle contour salient feature map and the semantic information such as structure in the essential feature map can be fully utilized to form a more comprehensive and rich feature representation. In addition, the attention-driven feature joint saliency perception network can strengthen the key information in the obstacle contour salient feature map and the obstacle essential feature map through the dependency matrix, while removing redundant parts, further improving the feature quality, and improving the model's sensitivity and generalization ability to input data, making the model more accurate and efficient when performing obstacle type recognition and detection tasks.

[0063] In one embodiment of the present application, the obstacle multi-scale feature joint perception module is used to: input the obstacle contour salient feature map and the obstacle essential feature map into an attention-driven feature joint saliency perception network to obtain the obstacle multi-scale salient joint perception feature map.

[0064] The obstacle multi-scale feature joint perception module includes: a first feature shape reshaping unit, used to perform feature shape reshaping on the obstacle contour salient feature map and the obstacle essential feature map to obtain an obstacle contour salient feature shape reshaping matrix and an obstacle essential feature shape reshaping matrix; a feature channel-by-channel interactive perception unit, used to input the obstacle contour salient feature shape reshaping matrix and the obstacle essential feature shape reshaping matrix into the feature channel-by-channel interactive perception module to obtain an obstacle contour-essential feature dependency matrix and an obstacle essence-contour feature dependency matrix; a random deactivation unit, used to input the obstacle contour-essential feature dependency matrix and the obstacle essence-contour feature dependency matrix into the random deactivation module to obtain a pruning optimization obstacle contour-essential feature dependency matrix and a pruning optimization obstacle essence-contour feature dependency matrix; a first matrix multiplication calculation unit, used to input the obstacle contour-essential feature dependency matrix and the obstacle essence-contour feature dependency matrix into the random deactivation module to obtain a pruning optimization obstacle contour-essential feature dependency matrix and a pruning optimization obstacle essence-contour feature dependency matrix; The pruned optimized obstacle contour-essential feature dependency matrix is ​​matrix-multiplied with the obstacle essential feature shape reshaping matrix to obtain a dependency optimized obstacle essential-contour feature dependency matrix; a second matrix multiplication calculation unit is used to multiply the pruned optimized obstacle essential-contour feature dependency matrix with the obstacle contour salient feature shape reshaping matrix to obtain a dependency optimized obstacle contour-essential feature dependency matrix; a second feature shape reshaping unit is used to perform feature shape reshaping on the dependency optimized obstacle essential-contour feature dependency matrix and the dependency optimized obstacle contour-essential feature dependency matrix to obtain an optimized obstacle contour salient feature map and an optimized obstacle essential feature map; a weighted sum calculation unit is used to calculate the weighted sum of the optimized obstacle contour salient feature map and the optimized obstacle essential feature map to obtain the obstacle multi-scale salient joint perception feature map.

[0065] Furthermore, in one embodiment of the present application, the feature channel-by-channel interaction perception unit is used to: perform matrix multiplication on the obstacle contour significant feature shape reshaping matrix and the transposed matrix of the obstacle essential feature shape reshaping matrix, and then divide the matrix by the scale of the obstacle contour-essential feature dependency matrix by position to obtain an obstacle contour-essential feature interaction matrix; perform matrix multiplication on the obstacle contour significant feature shape reshaping matrix and the obstacle essential feature shape reshaping matrix, and then divide the matrix by the scale of the obstacle essential feature shape reshaping matrix by position to obtain an obstacle essential-contour feature interaction matrix; use The obstacle contour-essential feature interaction matrix and the obstacle essence-contour feature interaction matrix are respectively normalized to obtain the obstacle contour-essential feature dependency matrix and the obstacle essence-contour feature dependency matrix.

[0066] Specifically, the obstacle contour salient feature map and the obstacle essential feature map are input into the attention-driven feature joint saliency perception network and processed by the following salient joint perception formula to obtain the obstacle multi-scale salient joint perception feature map;

[0067] Among them, the significant joint perception formula is:

[0068]

[0069]

[0070]

[0071]

[0072]

[0073]

[0074]

[0075]

[0076]

[0077]

[0078]

[0079] in, is the obstacle contour salient feature map, is the essential characteristic map of the obstacle, represents the feature shape reshaping process, and They are the shape reshaping matrix of the significant features of the obstacle contour and the shape reshaping matrix of the essential features of the obstacle, is the transpose operation, is the scale of the obstacle contour significant feature shape reshaping matrix and the obstacle essential feature shape reshaping matrix, for function, is matrix multiplication, and They are the obstacle contour-essential feature dependency matrix and the obstacle essence-contour feature dependency matrix, represents pruning optimization processing, and They are the pruning optimization obstacle contour-essential feature dependency matrix and the pruning optimization obstacle essence-contour feature dependency matrix, and They are respectively the dependency optimization obstacle contour-essential feature dependency matrix and the dependency optimization obstacle essence-contour feature dependency matrix. and They are respectively used to optimize the obstacle contour salient feature map and the obstacle essential feature map. and are all weighted hyperparameters, It is the multi-scale significant joint perception feature map of the obstacle.

[0080] Then, the obstacle multi-scale significant joint perception feature map is input into the recognition module based on the classifier to obtain the recognition result, and the recognition result is used to represent the type label of the obstacle. That is, the multi-scale feature significant joint representation information of the obstacle is used for classification processing to identify the type of obstacle, and based on the recognition result, it is determined whether to generate an alarm signal, which can use artificial intelligence and deep learning technology to more accurately identify and detect obstacles between platform doors and trains, and distinguish different types of obstacles, thereby reducing the possibility of false alarms and missed alarms, and realizing more intelligent safety protection of urban rail platform doors. In addition, this intelligent recognition and alarm method can also maintain a high obstacle type recognition accuracy in complex environments, which helps to improve the reliability and stability of the entire system.

[0081] In one embodiment of the present application, the obstacle detection alarm module is used to: input the obstacle multi-scale significant joint perception feature map into a classifier-based recognition module to obtain a recognition result, wherein the recognition result is used to represent the type label of the obstacle; based on the recognition result, determine whether to generate an alarm signal.

[0082] In a preferred example, considering that the obstacle contour salient feature map and the obstacle essential feature map respectively represent the associated saliency features of the contour image semantics of the three-dimensional scanning model of the obstacle based on the image semantic feature gradient distribution and the image semantic features of the backbone image semantics of the three-dimensional scanning model of the obstacle, when performing attention-driven feature joint saliency perception on them, the difference in attention weights caused by the difference in the spatial domain expression dimensions of the image semantic features will introduce image semantic joint perception aggregation offset. Therefore, it is expected to further improve the semantic spatial domain aggregation regression comprehensibility of the obstacle multi-scale salient joint perception feature map, thereby improving the accuracy of the recognition result obtained by inputting it into the classifier-based recognition module.

[0083] Based on this, in the preferred example, inputting the obstacle multi-scale significant joint perception feature map into a classifier-based recognition module to obtain a recognition result specifically includes:

[0084] Expanding the obstacle multi-scale significant joint perception feature map into an obstacle multi-scale significant joint perception feature vector;

[0085] Subtracting the zero norm of the obstacle multi-scale significant joint perception feature vector from the length of the obstacle multi-scale significant joint perception feature vector to obtain an obstacle multi-scale significant joint perception isolated zero value;

[0086] Calculating the logarithm of the square of the obstacle multi-scale significant joint perception isolated zero value and the sum of the obstacle multi-scale significant joint perception isolated zero value to obtain the obstacle multi-scale significant joint perception superposition information value;

[0087] Calculate a power function with each eigenvalue of the obstacle multi-scale significant joint perception eigenvector as the base, the difference of the obstacle multi-scale significant joint perception isolated zero value minus one as the exponent, and perform a point multiplication with the obstacle multi-scale significant joint perception superposition information value to obtain a first obstacle multi-scale significant joint perception intermediate vector;

[0088] After performing a dot multiplication on the obstacle multi-scale significant joint perception feature vector and the difference of the obstacle multi-scale significant joint perception isolated zero value minus one, the dot multiplication is performed on the inverse of the obstacle multi-scale significant joint perception isolated zero value to obtain a second obstacle multi-scale significant joint perception intermediate vector;

[0089] Calculate an exponential function with a natural constant as the base and each eigenvalue of the second obstacle multi-scale significant joint perception intermediate vector as the exponent to obtain a third obstacle multi-scale significant joint perception intermediate vector;

[0090] performing dot addition of the first obstacle multi-scale significant joint perception intermediate vector and the third obstacle multi-scale significant joint perception intermediate vector to obtain an optimized obstacle multi-scale significant joint perception feature vector;

[0091] The optimized obstacle multi-scale significant joint perception feature vector is input into a classifier-based recognition module to obtain a recognition result.

[0092] That is, the obstacle multi-scale significant joint perception feature vector, for example, is recorded as The optimization process is expressed as:

[0093]

[0094]

[0095] in, is the multi-scale significant joint perception feature vector of the obstacle, is the zero norm of the vector, is the length of the obstacle multi-scale significant joint perception feature vector, is the isolated zero value of multi-scale significant joint perception of obstacles, is the base-two logarithm, represents the exponential operation of a vector, wherein the exponential operation of a vector represents the calculation of the value of a natural exponential function raised to the power of the eigenvalues ​​at each position in the vector, is the optimized obstacle multi-scale significant joint perception feature vector, It is the point multiplication by position. It is positional addition.

[0096] Specifically, the high-dimensional feature manifold of the obstacle multi-scale significant joint perception feature map is represented by the feature value of the feature set as the vector field of the aggregation dimension, the superposition value of the vector field of the obstacle multi-scale significant joint perception feature map at the isolated zero position is used as the order information, the local position of the feature value of the feature set is fixed, and a bias for the reversibility of the feature regression distribution field of the obstacle multi-scale significant joint perception feature map is added as a reward to achieve the mapping target tracking of the regression distribution of the obstacle multi-scale significant joint perception feature map for the feature value position, so that the feature set of the obstacle multi-scale significant joint perception feature map perceives the mapping migration to the aggregate distribution, thereby improving the semantic space domain aggregation regression comprehensibility of the obstacle multi-scale significant joint perception feature map, improving the accuracy of the recognition result obtained by the obstacle multi-scale significant joint perception feature map input based on the classifier recognition module, and being able to more accurately identify and detect obstacles between the platform door and the train, and make corresponding alarms based on the recognition results, thereby reducing the possibility of false alarms and missed alarms, and realizing more intelligent urban railway platform door safety protection.

[0097] In summary, the above scheme uses a laser radar to collect a three-dimensional scanning model of an obstacle, and then uses a data processing algorithm based on artificial intelligence and deep learning to analyze the three-dimensional scanning model at the back end, so as to capture the outline and essential joint feature representation of the obstacle, and use this feature representation to identify the type of obstacle, so as to make a corresponding alarm based on the recognition result, and can use artificial intelligence and deep learning technology to more accurately identify and detect obstacles between platform doors and trains, and distinguish different types of obstacles, thereby reducing the possibility of false alarms and missed alarms, and realizing more intelligent safety protection for urban rail platform doors. In addition, this intelligent recognition and alarm method can also maintain a high obstacle type recognition accuracy in complex environments, which helps to improve the reliability and stability of the entire system.

[0098] Figure 2 is a flow chart of a method for protecting a platform door of an urban railway based on laser detection according to an exemplary embodiment. Figure 2 As shown, the method includes:

[0099] Step S201, obtaining a three-dimensional scanning model of an obstacle collected by a laser radar;

[0100] Step S202: input the three-dimensional scanning model of the obstacle into a double-branch boundary information loss compensation network including a boundary feature extraction branch and a backbone network to obtain an obstacle contour feature map and an obstacle essential feature map;

[0101] Step S203, inputting the obstacle contour feature map into a mask salient device based on feature gradient correlation distribution to obtain an obstacle contour salient feature map;

[0102] Step S204, performing feature joint saliency semantic perception on the obstacle contour saliency feature map and the obstacle essential feature map to obtain an obstacle multi-scale saliency joint perception feature map;

[0103] Step S205: Determine the type label of the obstacle based on the obstacle multi-scale significant joint perception feature map, and determine whether to generate an alarm signal.

[0104] In one embodiment of the present application, the obstacle contour feature map is input into a mask salient device based on feature gradient association distribution to obtain an obstacle contour salient feature map, including: calculating the multi-directional gradient value distribution of each position in the obstacle contour feature map, and determining the gradient amplitude value of each position in the obstacle contour feature map based on the multi-directional gradient value distribution of each position to obtain an obstacle contour feature gradient amplitude distribution feature map; calculating the gradient amplitude local description operator of each position in the obstacle contour feature gradient amplitude distribution feature map to obtain an obstacle contour feature gradient amplitude local significant distribution feature map; inputting the obstacle contour feature gradient amplitude local significant distribution feature map into a gated masker based on a GELU function to obtain an obstacle contour feature gradient amplitude local significant gated mask feature map; calculating the position point multiplication between the obstacle contour feature gradient amplitude local significant gated mask feature map and the obstacle contour feature map to obtain the obstacle contour salient feature map.

[0105] Reference below Figure 3 , which shows a schematic diagram of the structure of an electronic device 600 suitable for implementing the embodiment of the present application. The terminal device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 3The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0106] like Figure 3 As shown, the electronic device 600 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device 600 are also stored. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0107] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device 600 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 3 The electronic device 600 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0108] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 609, or installed from the storage device 608, or installed from the ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the method of the embodiment of the present application are executed.

[0109] It should be noted that the computer-readable medium mentioned above in the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable medium mentioned above may be included in the electronic device mentioned above; or it may exist independently without being assembled into the electronic device.

[0110] In the context of the present application, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0111] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

[0112] In addition, although each operation is described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or to be performed in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the application. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0113] Although the subject matter has been described in language specific to structural features and / or method logic actions, it should be understood that the subject matter defined is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely exemplary forms of implementation. With respect to the apparatus in the above-mentioned embodiments, the specific manner in which each module performs the operation has been described in detail in the embodiments related to the method, and will not be described in detail here.

Claims

1. A safety protection system for urban railway platform doors based on laser detection, characterized in that: include: The obstacle 3D scanning model acquisition module is used to obtain the 3D scanning model of the obstacle collected by the laser radar; An obstacle feature extraction module, used for inputting the three-dimensional scanning model of the obstacle into a double-branch boundary information loss compensation network including a boundary feature extraction branch and a trunk network to obtain an obstacle contour feature map and an obstacle essential feature map; A feature gradient correlation feature saliency enhancement module, used for inputting the obstacle contour feature map into a mask salient device based on feature gradient correlation distribution to obtain an obstacle contour salient feature map; An obstacle multi-scale feature joint perception module, used for performing feature joint significant semantic perception on the obstacle contour significant feature map and the obstacle essential feature map to obtain an obstacle multi-scale significant joint perception feature map; The obstacle detection alarm module is used to determine the type label of the obstacle based on the multi-scale significant joint perception feature map of the obstacle, and determine whether to generate an alarm signal.

2. According to the laser detection-based urban railway platform door safety protection system of claim 1, it is characterized in that: The feature gradient associated feature significance enhancement module comprises: a multi-directional gradient value distribution calculation unit, configured to calculate the multi-directional gradient value distribution of each position in the obstacle contour feature map, and determine the gradient amplitude value of each position in the obstacle contour feature map based on the multi-directional gradient value distribution of each position to obtain an obstacle contour feature gradient amplitude distribution feature map; A gradient amplitude local description operator calculation unit, used to calculate the gradient amplitude local description operator of each position in the obstacle contour feature gradient amplitude distribution feature map to obtain the obstacle contour feature gradient amplitude local significant distribution feature map; A gated mask calculation unit, used for inputting the obstacle contour feature gradient amplitude local significant distribution feature map into a gated masker based on a GELU function to obtain an obstacle contour feature gradient amplitude local significant gated mask feature map; The feature map position point multiplication calculation unit is used to calculate the position point multiplication between the obstacle contour feature gradient amplitude local significant gated mask feature map and the obstacle contour feature map to obtain the obstacle contour significant feature map.

3. The urban railway platform door safety protection system based on laser detection according to claim 2 is characterized in that: The multi-directional gradient value distribution calculation unit is used to: Calculating the gradient amplitude value of each position feature value in the obstacle profile feature map along the height direction to obtain an obstacle profile feature height gradient distribution feature map; Calculating the gradient amplitude value along the width direction of each position feature value in the obstacle contour feature map to obtain an obstacle contour feature width gradient distribution feature map; Calculating the gradient amplitude value of each position feature value in the obstacle profile feature map along the channel direction to obtain an obstacle profile feature channel gradient distribution feature map; The binary norm between the characteristic values ​​at corresponding positions in the obstacle contour feature height gradient distribution characteristic map, the obstacle contour feature width gradient distribution characteristic map and the obstacle contour feature channel gradient distribution characteristic map is calculated to obtain the obstacle contour feature gradient amplitude distribution characteristic map.

4. The urban railway platform door safety protection system based on laser detection according to claim 3 is characterized in that: The gradient amplitude local description operator calculation unit is used to: Determine a local neighborhood range of gradient amplitude, and calculate a sequence of obstacle contour feature gradient amplitude distribution neighborhood feature maps based on the local neighborhood range of gradient amplitude and the obstacle contour feature gradient amplitude distribution feature map; Counting the number of feature maps in the sequence of the obstacle contour feature gradient amplitude distribution neighborhood feature maps to obtain a statistical value of the number of obstacle contour feature gradient amplitude distribution neighborhoods; Calculating the position difference between each obstacle contour feature gradient amplitude distribution neighborhood feature map in the sequence of the obstacle contour feature gradient amplitude distribution feature map and the obstacle contour feature gradient amplitude distribution neighborhood feature map, and dividing the difference by position point with the obstacle contour feature gradient amplitude distribution feature map to obtain a sequence of obstacle contour feature gradient relative amplitude distribution neighborhood feature maps; Each obstacle contour feature gradient relative amplitude distribution neighborhood feature map in the sequence of the obstacle contour feature gradient relative amplitude distribution neighborhood feature map is added point by point and then divided by the obstacle contour feature gradient amplitude distribution neighborhood number statistics value point by point to obtain the obstacle contour feature gradient amplitude local significant distribution feature map.

5. The urban railway platform door safety protection system based on laser detection according to claim 4 is characterized in that: The obstacle multi-scale feature joint perception module is used to: input the obstacle contour salient feature map and the obstacle essential feature map into an attention-driven feature joint saliency perception network to obtain the obstacle multi-scale salient joint perception feature map.

6. The urban railway platform door safety protection system based on laser detection according to claim 5 is characterized in that: The obstacle multi-scale feature joint perception module includes: A first characteristic shape reshaping unit is used to perform characteristic shape reshaping on the obstacle contour significant feature map and the obstacle essential feature map to obtain an obstacle contour significant feature shape reshaping matrix and an obstacle essential feature shape reshaping matrix; A feature channel-by-channel interactive perception unit, used for inputting the obstacle contour salient feature shape reshaping matrix and the obstacle essential feature shape reshaping matrix into a feature channel-by-channel interactive perception module to obtain an obstacle contour-essential feature dependency matrix and an obstacle essence-contour feature dependency matrix; A random deactivation unit, used for inputting the obstacle contour-essential feature dependency matrix and the obstacle essence-contour feature dependency matrix into a random deactivation module to obtain a pruned optimized obstacle contour-essential feature dependency matrix and a pruned optimized obstacle essence-contour feature dependency matrix; a first matrix multiplication calculation unit, configured to perform matrix multiplication on the pruning optimization obstacle contour-essential feature dependency matrix and the obstacle essential feature shape reshaping matrix to obtain a dependency optimization obstacle essence-contour feature dependency matrix; A second matrix multiplication calculation unit is used to multiply the pruning optimization obstacle essence-contour feature dependency matrix with the obstacle contour significant feature shape reshaping matrix to obtain a dependency optimization obstacle contour-essential feature dependency matrix; A second characteristic shape reshaping unit is used to perform characteristic shape reshaping on the dependency-optimized obstacle essence-contour feature dependency matrix and the dependency-optimized obstacle contour-essential feature dependency matrix to obtain an optimized obstacle contour significant feature map and an optimized obstacle essential feature map; A weighted sum calculation unit is used to calculate the weighted sum of the optimized obstacle contour salient feature map and the optimized obstacle essential feature map to obtain the obstacle multi-scale salient joint perception feature map.

7. The urban railway platform door safety protection system based on laser detection according to claim 6 is characterized in that: The feature channel-by-channel interactive perception unit is used to: The obstacle contour-essential feature interaction matrix is ​​obtained by performing matrix multiplication of the transposed matrix of the obstacle contour salient feature shape reshaping matrix and the obstacle essential feature shape reshaping matrix, and then dividing the matrix by the scale of the obstacle contour-essential feature dependency matrix by position; The transposed matrix of the obstacle contour salient feature shape reshaping matrix and the obstacle essential feature shape reshaping matrix are matrix-multiplied, and then the matrix is ​​scaled by dividing the matrix by the obstacle essential feature shape reshaping matrix by position to obtain an obstacle essential-contour feature interaction matrix; use The obstacle contour-essential feature interaction matrix and the obstacle essence-contour feature interaction matrix are respectively normalized to obtain the obstacle contour-essential feature dependency matrix and the obstacle essence-contour feature dependency matrix.

8. The urban railway platform door safety protection system based on laser detection according to claim 7 is characterized in that: The obstacle detection alarm module is used to: Inputting the obstacle multi-scale significant joint perception feature map into a classifier-based recognition module to obtain a recognition result, wherein the recognition result is used to represent a type label of the obstacle; Based on the recognition result, it is determined whether to generate an alarm signal.

9. A safety protection method for urban railway platform doors based on laser detection, characterized in that: include: Obtain a three-dimensional scanning model of the obstacle collected by the laser radar; Inputting the three-dimensional scanning model of the obstacle into a double-branch boundary information loss compensation network including a boundary feature extraction branch and a backbone network to obtain an obstacle contour feature map and an obstacle essential feature map; Inputting the obstacle contour feature map into a mask salient device based on feature gradient correlation distribution to obtain an obstacle contour salient feature map; Performing feature joint saliency semantic perception on the obstacle contour saliency feature map and the obstacle essential feature map to obtain an obstacle multi-scale saliency joint perception feature map; Based on the obstacle multi-scale significant joint perception feature map, the type label of the obstacle is determined, and it is determined whether to generate an alarm signal.

10. The method for protecting the platform door of an urban railway based on laser detection according to claim 9 is characterized in that: Inputting the obstacle contour feature map into a mask salient device based on feature gradient correlation distribution to obtain an obstacle contour salient feature map, comprising: Calculating the multi-directional gradient value distribution of each position in the obstacle contour feature map, and determining the gradient amplitude value of each position in the obstacle contour feature map based on the multi-directional gradient value distribution of each position to obtain an obstacle contour feature gradient amplitude distribution feature map; Calculating the local description operator of the gradient amplitude at each position in the obstacle contour feature gradient amplitude distribution feature map to obtain the obstacle contour feature gradient amplitude local significant distribution feature map; Inputting the obstacle contour feature gradient amplitude local significant distribution feature map into a gated masker based on a GELU function to obtain an obstacle contour feature gradient amplitude local significant gated mask feature map; The obstacle contour salient feature map is obtained by calculating the point-by-point multiplication between the obstacle contour feature gradient amplitude local salient gated mask feature map and the obstacle contour feature map.

Images