Environment sensing method of intelligent rail vehicle and related device
Through lidar and point cloud processing technology, combined with detection and segmentation algorithms, the problem of intelligent track sensing needs in harsh environments is solved, high-precision and real-time environmental perception are achieved, and the needs of intelligent tracking operation are met.
Patent Information
- Application Number
- CN202311440958.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-01
- Publication Date
- 2025-05-06
AI Technical Summary
The existing image detection technology cannot meet the perceived needs of the stable operation of the intelligent track in the entire period of time under the environment of strong light, heavy rain, and dark night.
Lidar is used for environmental perception, through data annotation and model training of point clouds, detection algorithms and segmentation algorithms are used to detect obstacles and segmentation point clouds, and real-time detection information and segmentation information are output.
It improves the distance detection accuracy and urban environmental adaptability, enhances the obstacle detection accuracy and real-time detection, and meets the perceived needs of intelligent orbits running all-weather.
Smart Images
Figure CN119942490A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of intelligent rail vehicle control technology, and in particular to an environment perception method for an intelligent rail vehicle; it also relates to an environment perception device, equipment and computer-readable storage medium for an intelligent rail vehicle. Background Art
[0002] The full name of Zhigui is Intelligent Rail Express System. Its special feature is that there is no track, but only guide markings drawn on the ground. The guide markings on the ground are identified by on-board optical instruments to achieve a fixed driving route like laying tracks. Zhigui is a new medium-capacity urban rail transit system that integrates autonomous guidance, track following, all-electric drive, and intelligent driving. Among them, environmental perception is required for the operation of Zhigui vehicles. However, the detection accuracy of existing image detection technology will be affected by strong light, heavy rain, and dark nights, and cannot meet the perception requirements of the stable operation of Zhigui at all times. Therefore, providing an environmental perception solution that can meet the requirements of the normal and stable operation of Zhigui for detection range, detection effect, and real-time performance has become a technical problem that needs to be solved urgently by those skilled in the art. Summary of the invention
[0003] The purpose of this application is to provide an environment perception method for a smart rail vehicle, which has higher distance detection accuracy, better adaptability to urban environments, and better obstacle detection accuracy. Another purpose of this application is to provide an environment perception device, equipment, and computer-readable storage medium for a smart rail vehicle, all of which have the above-mentioned technical effects.
[0004] In order to solve the above technical problems, the present application provides an environment perception method for a smart rail vehicle, comprising:
[0005] Annotate the point cloud to obtain annotation information; the annotation information includes 3D obstacle labels and categories of points in the point cloud;
[0006] The model is trained using the point cloud and the annotation information; the model includes a detection algorithm and a segmentation algorithm; the detection algorithm is used to detect 3D obstacles, and the segmentation algorithm is used to segment points in the point cloud;
[0007] The real-time point cloud collected by the LiDAR is input into the trained model for inference, and the detection information and segmentation information are output.
[0008] Optionally, the training a model using the point cloud and the annotation information includes:
[0009] Using the point cloud and the annotation information to train the detection algorithm and the segmentation algorithm in the model respectively;
[0010] Evaluating the trained model to obtain evaluation indicators;
[0011] If the evaluation index meets the standard, the detection algorithm and the segmentation algorithm in the model are jointly trained and / or quantized using the point cloud and the annotation information.
[0012] Optionally, also include:
[0013] The categories of obstacles and points output by the model during the evaluation process are visualized and rendered.
[0014] Optionally, the training a model using the point cloud and the annotation information includes:
[0015] The point cloud and the annotation information are input into the model, and training is performed using back propagation through the pytorch framework.
[0016] Optionally, the step of inputting the point cloud into the trained model for inference and outputting the detection information and the segmentation information further includes:
[0017] The points in the real-time point cloud are converted into a pseudo image under the bev perspective; the pseudo image is divided into grids of the bev perspective according to distance, and the points in each grid are regarded as a cluster; wherein, during the conversion process, points in a range that does not need to be detected and / or segmented are removed, invalid points are filtered, and a number of points are randomly retained in the cluster.
[0018] Optionally, also include:
[0019] After information organization, threshold filtering and abnormal result elimination processing are performed on the detection information and segmentation information, the detection results and segmentation results are output in the form of messages.
[0020] Optionally, the detection head of the model locates the detection target in an Anchor Free manner.
[0021] In order to solve the above technical problems, the present application also provides an environment perception device for a smart rail vehicle, comprising:
[0022] A labeling module is used to label the point cloud data to obtain labeling information; the labeling information includes 3D obstacle labels and categories of points in the point cloud;
[0023] A training module, used to train a model using the point cloud and the annotation information; the model includes a detection algorithm and a segmentation algorithm; the detection algorithm is used to detect 3D obstacles, and the segmentation algorithm is used to segment points in the point cloud;
[0024] The inference module is used to input the real-time point cloud collected by the laser radar into the trained model for inference, and output detection information and segmentation information.
[0025] In order to solve the above technical problems, the present application also provides an environment perception device for a smart rail vehicle, including:
[0026] Memory for storing computer programs;
[0027] A processor is used to implement the steps of the environmental perception method of the smart rail vehicle as described above when executing the computer program.
[0028] In order to solve the above technical problems, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the environmental perception method of the smart rail vehicle as described above are implemented.
[0029] The environmental perception method for a smart rail vehicle provided in the present application includes: performing data annotation on a point cloud to obtain annotation information; the annotation information includes a 3D obstacle label and a category of a point in the point cloud; training a model using the point cloud and the annotation information; the model includes a detection algorithm and a segmentation algorithm; the detection algorithm is used to detect 3D obstacles, and the segmentation algorithm is used to segment points in the point cloud; the real-time point cloud collected by a laser radar is input into the trained model for inference, and the detection information and the segmentation information are output.
[0030] It can be seen that the environmental perception method of the smart rail vehicle provided in this application uses laser radar for environmental perception, has higher distance detection accuracy, better adaptability to urban environments, can cover general driving conditions, and meet the all-weather operation requirements of the smart rail. In addition, the model includes a detection algorithm and a segmentation algorithm. The perception scheme based on the detection and segmentation multi-task model has better obstacle detection accuracy and better detection real-time performance, which can provide timely and effective surrounding environment information for the decision-making planning and control of the smart rail vehicle.
[0031] The environmental perception device, equipment and computer-readable storage medium for smart rail vehicles provided in this application all have the above-mentioned technical effects. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the prior art and the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0033] Figure 1 A schematic diagram of a flow chart of an environment perception method for a smart rail vehicle provided in an embodiment of the present application;
[0034] Figure 2A schematic diagram of an environment perception device for a smart rail vehicle provided in an embodiment of the present application;
[0035] Figure 3 A schematic diagram of an environmental perception device for a smart rail vehicle provided in an embodiment of the present application. DETAILED DESCRIPTION
[0036] The core of this application is to provide an environment perception method for a smart rail vehicle, which has higher distance detection accuracy, better adaptability to urban environments, and better obstacle detection accuracy. Another core of this application is to provide an environment perception device, equipment, and computer-readable storage medium for a smart rail vehicle, all of which have the above-mentioned technical effects.
[0037] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0038] Please refer to Figure 1 , Figure 1 A flow chart of an environment perception method for a smart rail vehicle provided in an embodiment of the present application, referring to Figure 1 As shown, the method includes:
[0039] S101: Annotate the point cloud to obtain annotation information; the annotation information includes 3D obstacle labels and categories of points in the point cloud;
[0040] S102: training a model using the point cloud and the annotation information; the model includes a detection algorithm and a segmentation algorithm; the detection algorithm is used to detect 3D obstacles, and the segmentation algorithm is used to segment points in the point cloud;
[0041] Steps S101 and S102 are intended to train the model offline. A point cloud is a collection of massive points collected by a lidar that reflects the surface characteristics of a target in three-dimensional space. The categories of 3D obstacles and each point in the point cloud can be framed on a dedicated annotation software. Specifically, after a frame of point cloud is collected, a 3D box is used to frame the points in the point cloud that belong to the same 3D obstacle, and these points are used as raw data for the detection task. Among the remaining points, if there are points belonging to the ground, fences, or smart rail platforms, they are marked out for segmentation tasks. After obtaining the annotation information, the annotation information is converted into a specific format and returned. The specific format can specifically be the KITTI dataset label annotation file format. The annotation information includes 3D obstacle labels and categories of points in the point cloud. 3D obstacle labels include obstacle categories, positions, and orientations.
[0042] The model is trained using the point cloud and annotation information. In this embodiment, the detection task and the segmentation task are implemented in the same network structure, and the model includes a detection algorithm and a segmentation algorithm. Training the model mainly involves training the detection algorithm and the segmentation algorithm.
[0043] In some embodiments, the training of the model using the point cloud and the annotation information includes:
[0044] Using the point cloud and the annotation information to train the detection algorithm and the segmentation algorithm in the model respectively;
[0045] Evaluating the trained model to obtain evaluation indicators;
[0046] If the evaluation index meets the standard, the detection algorithm and the segmentation algorithm in the model are jointly trained and / or quantized using the point cloud and the annotation information.
[0047] First, use the point cloud and annotation information to train the detection algorithm and the segmentation algorithm respectively. To train the detection algorithm, input the point cloud and 3D obstacle labels. To train the segmentation algorithm, input the point cloud and the categories of the points in the point cloud. After the training is completed, input the trained model and the point cloud of the test set, 3D obstacle labels, the categories of the points in the point cloud, and other annotation information into the model evaluation code. The model evaluation code evaluates by comparing the categories of the 3D obstacles output by the model and the points in the segmentation information point cloud to obtain the evaluation indicators. Segmentation information refers to the category to which each point in the point cloud belongs. If the overlap ratio between the 3D obstacle output by the model and the annotated 3D obstacle is greater than a certain value, the detection is judged to be correct. If the category of the point output by the model is consistent with the category of the annotated point, the segmentation of the point is correct.
[0048] For 3D obstacle detection, the evaluation metric is average accuracy. The evaluation metric for the entire model is the mean of the average accuracy of each category. For point cloud segmentation, the evaluation metrics are overall pixel accuracy, average accuracy of each category, and average intersection.
[0049] If the evaluation indicators (for example, the average accuracy of each category and the overall pixel accuracy) are not up to standard, continue to add badcase data (data with unsatisfactory model detection) for manual annotation to iterate the model. If the evaluation indicators are up to standard, joint training or quantitative training is performed. Among them, joint training refers to the joint training of the detection algorithm and the segmentation algorithm. At the same time, the point cloud, 3D obstacle labels, and the categories of the points in the point cloud are input into the model to train the detection algorithm and the segmentation algorithm. By adjusting the training strategy through joint training, a balanced effect between the two subtasks of detection and segmentation can be obtained according to needs, and quantitative training can effectively ensure the accuracy of the deployment end. The model can be evaluated again after joint training.
[0050] In some embodiments, the training of the model using the point cloud and the annotation information includes:
[0051] The point cloud and the annotation information are input into the model, and training is performed using back propagation through the pytorch framework.
[0052] After the 3D obstacle labels, the categories of the points in the point cloud, and the point cloud are input into the model, the pytorch framework is used for training using back propagation. After the training is completed, the pytorch pth model file is saved.
[0053] Pytorch is an open source Python-based machine learning library that is used to flexibly build deep learning models. It is clear that in addition to training through the Pytorch framework, it can also be trained through the TensorFlow framework.
[0054] In some embodiments, it also includes:
[0055] The categories of obstacles and points output by the model during the evaluation process are visualized and rendered.
[0056] The visualization code performs visualization rendering based on the categories of obstacles and points inferred in each frame output by the evaluation process, which makes it easier for relevant personnel to understand the reasoning results of the model and optimize the model.
[0057] Finally, after inputting the Pytorch trained PTH model into the model export module, the Pytorch PTH model is converted into an onnx intermediate format model through the Pytorch API, and then the onnx model is converted into a TensorRT engine model using the TensorRT tool.
[0058] S103: Input the real-time point cloud acquired by the laser radar into the trained model for inference, and output the detection result and the segmentation result.
[0059] Step S103 is to use the trained model for online detection. The trained model is deployed to the smart rail vehicle, and the real-time point cloud collected by the laser radar on the smart rail vehicle is input into the trained model. After the model is inferred, the obstacle information, i.e., detection information and segmentation information, is output, so as to perceive the 3D obstacles in the surrounding environment and divide the drivable area. The scenes that can be perceived cover all scenes in the smart rail operation line, including but not limited to intersection scenes, platform entry and exit scenes, etc.
[0060] In some embodiments, the step of inputting the point cloud into the trained model for inference and outputting the detection information and the segmentation information further comprises:
[0061] The points in the real-time point cloud are converted into a pseudo image under the bev perspective; the pseudo image is divided into grids of the bev perspective according to distance, and the points in each grid are regarded as a cluster; wherein, during the conversion process, points in a range that does not need to be detected and / or segmented are removed, invalid points are filtered, and a number of points are randomly retained in the cluster.
[0062] Specifically, the online system is triggered to execute through the point cloud message in the roc, and pre-processing is performed before reasoning. Among them, the pre-processing mainly performs point cloud coordinate conversion, converting the points in the point cloud coordinate system into a pseudo-image with the radar as the origin, the forward direction as x, and the left direction as y under the bev perspective. The pseudo-image is divided into grids of bev perspective according to the distance, and the points in each grid are regarded as a cluster. During the conversion process, the points that do not need to be detected or segmented are removed, and then the invalid point cloud is filtered and the point cloud is screened in the clusters in the area. Point cloud screening is to randomly select n points in the cluster and retain them, and filter out the rest.
[0063] The cluster division method based on this embodiment and the feature mapping of the bev space can effectively improve the point cloud processing speed of the algorithm and meet the hardware deployment requirements. The full name of bev is Bird's Eye View, which refers to the projection of the point cloud on a plane perpendicular to the height direction.
[0064] In some embodiments, it also includes:
[0065] After information organization, threshold filtering and abnormal result elimination processing are performed on the detection information and segmentation information, the detection results and segmentation results are output in the form of messages.
[0066] After the reasoning is completed, post-processing is performed. Post-processing is to organize information, filter thresholds, and remove abnormal results based on the outputs of the detection head and the segmentation head, and finally output them in the form of messages.
[0067] The model discretizes the point cloud into evenly spaced grids on the xy plane, generating a series of Pillars (a three-dimensional small cell obtained by dividing the point cloud with a certain step size on the XY plane of the point cloud space (Cartesian coordinate system)), and then performs local feature extraction on each pillar, and then transforms the extracted features into the bev space through mapping. Afterwards, based on the classic feature pyramid form, features are first extracted from top to bottom, and finally each dimensional feature is upsampled to obtain features of different scales, and finally multi-scale features are combined by splicing. The last part is the detection head and segmentation head. In some embodiments, the detection head of the model uses Anchor Free to locate the detection target, including predicting key points such as corner points, extreme points, and center points. Specifically, the detection head uses Anchor Free to mainly regress Heatmap, center point Offset, Heading angle, Size, and Z. Among them, Heatmap represents the activation of the center point of each type of obstacle, the center point Offset is the offset of the obstacle relative to the center of the Pillar, the Heading angle reflects the direction of the obstacle, Size is the length, width and height of the corresponding object, and Z is the height information of the center point of the corresponding object. The segmentation head uses a general 3x3 convolution method to predict pixel-level segmentation information.
[0068] In summary, the environmental perception method of the smart rail vehicle provided in this application uses laser radar for environmental perception, has higher distance detection accuracy, better adaptability to urban environments, can cover general driving conditions, and meet the all-weather operation requirements of the smart rail. In addition, the model includes a detection algorithm and a segmentation algorithm. The perception scheme based on the detection and segmentation multi-task model has better obstacle detection accuracy and better detection real-time performance, which can provide timely and effective surrounding environment information for the decision-making, planning and control of the smart rail vehicle.
[0069] The present application also provides an environment perception device for a smart rail vehicle. The device described below can be referred to in correspondence with the method described above. Figure 2 , Figure 2 A schematic diagram of an environment sensing device for a smart rail vehicle provided in an embodiment of the present application, combined with Figure 2 As shown, the device comprises:
[0070] The annotation module 10 is used to annotate the point cloud data to obtain annotation information; the annotation information includes 3D obstacle labels and categories of points in the point cloud;
[0071] A training module 20, for training a model using the point cloud and the annotation information; the model includes a detection algorithm and a segmentation algorithm; the detection algorithm is used to detect 3D obstacles, and the segmentation algorithm is used to segment points in the point cloud;
[0072] The reasoning module 30 is used to input the real-time point cloud collected by the laser radar into the trained model for reasoning, and output detection information and segmentation information.
[0073] Based on the above embodiment, as a specific implementation, the training module 20 includes:
[0074] A first training unit, used to train the detection algorithm and the segmentation algorithm in the model using the point cloud and the annotation information respectively;
[0075] An evaluation unit, used to evaluate the trained model to obtain evaluation indicators;
[0076] The second training unit is used to use the point cloud and the annotation information to perform joint training and / or quantitative training on the detection algorithm and the segmentation algorithm in the model if the evaluation index meets the standard.
[0077] Based on the above embodiment, as a specific implementation method, it also includes:
[0078] The visualization module is used to visualize the categories of obstacles and points output by the model during the evaluation process.
[0079] Based on the above embodiment, as a specific implementation, the training module 20 is specifically used for:
[0080] The point cloud and the annotation information are input into the model, and training is performed using back propagation through the pytorch framework.
[0081] Based on the above embodiment, as a specific implementation method, it also includes:
[0082] A pre-processing module is used to convert the points in the real-time point cloud into a pseudo image under the bev perspective; the pseudo image is divided into grids of the bev perspective according to distance, and the points in each grid are regarded as a cluster; wherein, during the conversion process, points in a range that does not need to be detected and / or segmented are removed, invalid points are filtered, and a number of points are randomly retained in the cluster.
[0083] Based on the above embodiment, as a specific implementation method, it also includes:
[0084] The post-processing module is used to organize the detection information and the segmentation information, perform threshold filtering and remove abnormal results, and then output the detection results and the segmentation results in the form of messages.
[0085] Based on the above embodiment, as a specific implementation mode, the detection head of the model uses Anchor Free to locate the detection target.
[0086] This application also provides an environment perception device for a smart rail vehicle, referring to Figure 3 As shown, the device includes a memory 1 and a processor 2 .
[0087] Memory 1, used for storing computer programs;
[0088] Processor 2 is used to execute the computer program to implement the following steps:
[0089] The point cloud is annotated to obtain annotation information; the annotation information includes 3D obstacle labels and categories of points in the point cloud; the model is trained using the point cloud and the annotation information; the model includes a detection algorithm and a segmentation algorithm; the detection algorithm is used to detect 3D obstacles, and the segmentation algorithm is used to segment points in the point cloud; the real-time point cloud collected by the laser radar is input into the trained model for inference, and the detection information and segmentation information are output.
[0090] For an introduction to the equipment provided in this application, please refer to the above method embodiments, and this application will not go into details here.
[0091] The present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps can be implemented:
[0092] The point cloud is annotated to obtain annotation information; the annotation information includes 3D obstacle labels and categories of points in the point cloud; the model is trained using the point cloud and the annotation information; the model includes a detection algorithm and a segmentation algorithm; the detection algorithm is used to detect 3D obstacles, and the segmentation algorithm is used to segment points in the point cloud; the real-time point cloud collected by the laser radar is input into the trained model for inference, and the detection information and segmentation information are output.
[0093] The computer-readable storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.
[0094] For an introduction to the computer-readable storage medium provided in this application, please refer to the above method embodiment, and this application will not go into details here.
[0095] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the devices, equipment, and computer-readable storage media disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part description.
[0096] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0097] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
[0098] The above is a detailed introduction to the environmental perception method, device, equipment and computer-readable storage medium for the smart rail vehicle provided by the present application. This article uses specific examples to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of the present application.
Claims
1. A method for environmental perception of a smart rail vehicle, characterized in that: include: Annotate the point cloud data to obtain annotation information; The annotation information includes 3D obstacle labels and categories of points in the point cloud; The model is trained using the point cloud and the annotation information; the model includes a detection algorithm and a segmentation algorithm; the detection algorithm is used to detect 3D obstacles, and the segmentation algorithm is used to segment points in the point cloud; The real-time point cloud collected by the LiDAR is input into the trained model for inference, and the detection information and segmentation information are output.
2. The environment perception method according to claim 1, characterized in that: The training of the model using the point cloud and the annotation information includes: Using the point cloud and the annotation information to train the detection algorithm and the segmentation algorithm in the model respectively; Evaluating the trained model to obtain evaluation indicators; If the evaluation index meets the standard, the detection algorithm and the segmentation algorithm in the model are jointly trained and / or quantized using the point cloud and the annotation information.
3. The environment perception method according to claim 2, characterized in that: Also includes: The categories of obstacles and points output by the model during the evaluation process are visualized and rendered.
4. The environment perception method according to claim 1, characterized in that: The training of the model using the point cloud and the annotation information includes: The point cloud and the annotation information are input into the model, and training is performed using back propagation through the pytorch framework.
5. The environment perception method according to claim 1, characterized in that: The step of inputting the point cloud into the trained model for inference and outputting the detection information and the segmentation information also includes: The points in the real-time point cloud are converted into a pseudo image under the bev perspective; the pseudo image is divided into grids of the bev perspective according to distance, and the points in each grid are regarded as a cluster; wherein, during the conversion process, points in a range that does not need to be detected and / or segmented are removed, invalid points are filtered, and a number of points are randomly retained in the cluster.
6. The environment perception method according to claim 5, characterized in that: Also includes: After information organization, threshold filtering and abnormal result elimination processing are performed on the detection information and segmentation information, the detection results and segmentation results are output in the form of messages.
7. The environment perception method according to claim 1, characterized in that: The detection head of the model uses AnchorFree to locate the detection target.
8. An environment perception device for a smart rail vehicle, characterized in that: include: The annotation module is used to annotate the point cloud data and obtain annotation information; The annotation information includes 3D obstacle labels and categories of points in the point cloud; A training module, used to train a model using the point cloud and the annotation information; the model includes a detection algorithm and a segmentation algorithm; the detection algorithm is used to detect 3D obstacles, and the segmentation algorithm is used to segment points in the point cloud; The inference module is used to input the real-time point cloud collected by the laser radar into the trained model for inference, and output detection information and segmentation information.
9. An environment perception device for a smart rail vehicle, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the environment perception method for a smart rail vehicle as described in any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the environmental perception method for a smart rail vehicle as described in any one of claims 1 to 7 are implemented.