Train position initialization method, device, equipment and medium

Abstract

The invention provides a train position initialization method, device and equipment and a medium, and the method comprises the steps: obtaining peripheral images and point cloud data of each point location through a vehicle-mounted image collection device and a laser radar sensor according to a plurality of point locations preset in a rail-mounted area when a train is located at each point location; target semantic features are extracted through a machine learning algorithm based on the collected images; determining the relative position information of the train corresponding to each target semantic feature according to the internal and external parameter relationship between the laser radar and the sensor of the vehicle-mounted image acquisition device and the acquired point cloud data, and obtaining the semantic information of each point location; and the position information of the point locations in the global map and the corresponding semantic information of the point locations are used as position comparison information to realize train position initialization. According to the method, the position initialization of the trains of different signal systems is allowed to be realized on the same line, and the position initialization of the system trains with poor signals or even without signals can also be realized, so that the usability and the safety of train operation are effectively ensured.

Description

Technical Field

[0001] The present invention relates to the field of rail transit technology, and in particular to a train position initialization method, device, equipment and medium. Background Art

[0002] In the prior art, train control systems rely on transponders to achieve train positioning correction. Although this method is reliable, it will fail for trains with different signal systems (such as cross-line trains) or trains whose own signal systems are in a faulty state or lack signal equipment (such as engineering vehicles).

[0003] With the development of SLAM (simultaneous localization and mapping) technology based on optical sensors (lidar or camera), vehicle positioning technology based on images or lidar point clouds has become increasingly mature, making it possible to achieve train positioning that does not rely on traditional signal systems. However, SLAM must ensure that the vehicle updates its position from a known point. For urban rail transit lines, the GNSS signal in tunnels, elevated sections and other sections is poor, and it is difficult to obtain the initial position of the train through satellite positioning. The tracks and surrounding environments in tunnels, elevated sections and other sections have high repetitiveness, and it is difficult to ensure the accuracy of vehicle positioning initialization by simply matching non-semantic environmental features. Summary of the invention

[0004] In order to overcome the problems existing in the related art, the present disclosure provides a train position initialization method, device, equipment and medium to solve the technical problems in the related art.

[0005] One or more embodiments of this specification provide a method for initializing a train position, comprising the following steps:

[0006] According to multiple preset points in the track area, when the train is located at each point, the image and point cloud data around each point are obtained through the on-board image acquisition device and the laser radar sensor;

[0007] Extracting target semantic features based on the collected images through machine learning algorithms;

[0008] According to the relationship between the internal and external parameters of the laser radar and the on-board image acquisition device, and the collected point cloud data, the relative position information of the train corresponding to each target semantic feature is determined to obtain the semantic information of each point; and

[0009] The train position is initialized using the location information of the point in the global map and the corresponding semantic information of the point as position comparison information.

[0010] Preferably, the point is the location of a landmark with significant features; or

[0011] The train arrives at the position of the ground track at preset time intervals;

[0012] The image includes an object with a fixed position.

[0013] Preferably, the machine learning algorithm is a target detection algorithm or an instance segmentation algorithm, which can segment target features in an image and determine the semantics corresponding to each target feature.

[0014] Preferably, the train position initialization is further implemented using the position information of the point in the global map and the corresponding point semantic information as position comparison information, which specifically includes the following steps:

[0015] Construct a semantic-location dictionary based on the location information of the point in the global map and the corresponding semantic information of the point;

[0016] When the train needs to implement position initialization, the surrounding images and point cloud data of the current position are collected by the on-board image acquisition device and the laser radar respectively, the image target semantic features and the relative position information of the train corresponding to each target semantic feature are determined, and the semantic information corresponding to the train position is determined; and

[0017] The obtained semantic information is compared with the semantic-position dictionary to obtain the position information as the train initialization position.

[0018] One or more embodiments of this specification provide a train position initialization device, including:

[0019] The data acquisition module is used to obtain the surrounding images and point cloud data of each point through the on-board image acquisition device and the laser radar sensor according to multiple points preset in the track area when the train is located at each point;

[0020] A feature extraction module is used to extract target semantic features based on the collected images through a machine learning algorithm;

[0021] A semantic information determination module is used to determine the relative position information of the train corresponding to each target semantic feature based on the relationship between the internal and external parameters of the sensors of the laser radar and the vehicle-mounted image acquisition device, and the collected point cloud data, and obtain the semantic information of each point; and

[0022] The position initialization module is used to initialize the train position using the position information of the point in the global map and the corresponding point semantic information as position comparison information.

[0023] Preferably, the point is the location of a landmark with significant features; or

[0024] The train arrives at the position of the ground track at preset time intervals;

[0025] The image includes an object with a fixed position.

[0026] Preferably, the machine learning algorithm is a target detection algorithm or an instance segmentation algorithm, which can segment target features in an image and determine the semantics corresponding to each target feature.

[0027] Preferably, the position initialization module is further configured to perform the following steps to implement train position initialization, and the specific steps include:

[0028] Construct a semantic-location dictionary based on the location information of the point in the global map and the corresponding semantic information of the point;

[0029] When the train needs to implement position initialization, the surrounding images and point cloud data of the current position are collected by the on-board image acquisition device and the laser radar respectively, the image target semantic features and the relative position information of the train corresponding to each target semantic feature are determined, and the semantic information corresponding to the train position is determined; and

[0030] The obtained semantic information is compared with the semantic-position dictionary to obtain the position information as the train initialization position.

[0031] One or more embodiments of the present specification provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the train position initialization method as described in any one of the above items when executing the computer program.

[0032] One or more embodiments of the present specification provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the train position initialization method as described in any one of the above items is implemented.

[0033] The present invention provides a method, device, equipment and medium for initializing train position, which has the advantage that images collected by a train image acquisition device and a laser radar point cloud are used as information sources, and there is no need to build infrastructure such as transponders of traditional rail transit signal systems. Position comparison information is obtained by combining the position information of each preset point in the global map, thereby effectively initializing the train position. This allows trains of different signal systems to achieve position initialization on the same line, and can also achieve position initialization for trains of systems with poor signals or even no signals, effectively ensuring the availability and safety of train operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate one or more embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 A flow chart of a train position initialization method provided for one or more embodiments of this specification;

[0036] Figure 2 One or more embodiments of this specification provide a schematic diagram of using a signal machine in a rail transit line as an initialization point;

[0037] Figure 3 A target semantic feature information graph in an image taken near a signal point provided by one or more embodiments of this specification;

[0038] Figure 4 One or more embodiments of this specification provide a Figure 3 A schematic diagram of the relative spatial position relationship of the target semantic features in the determined image;

[0039] Figure 5 One or more embodiments of this specification provide a schematic diagram of using a stop sign in a rail transit line as an initialization point;

[0040] Figure 6 A target semantic feature information map in an image taken near a parking mark point provided by one or more embodiments of this specification;

[0041] Figure 7 One or more embodiments of this specification provide a Figure 6 A schematic diagram of the relative spatial position relationship of the target semantic features in the determined image;

[0042] Figure 8 A block diagram of a train position initialization device provided for one or more embodiments of this specification; and

[0043] Fig. 9 A schematic diagram of the structure of a computer device provided for one or more embodiments of this specification. DETAILED DESCRIPTION

[0044] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below in conjunction with the drawings in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0045] The present invention is described in detail below in conjunction with specific implementation methods and the accompanying drawings.

[0046] Method Embodiment

[0047] According to an embodiment of the present invention, a train position initialization method is provided, such as Figure 1 As shown, it is a flow chart of the train position initialization method provided in this embodiment. The train position initialization method according to the embodiment of the present invention comprises the steps of:

[0048] Step S1, according to multiple points preset in the track area, when the train is located at each point, the image and point cloud data around the point are obtained through the on-board image acquisition device and the lidar sensor, wherein the image includes multiple objects with fixed positions.

[0049] In this embodiment, the point position may be a target position in a track line selected according to demand, and generally a landmark with significant features is used as the initialization point position, such as a signal light, a stop sign, or the position corresponding to other objects with fixed positions, or the position where a train arrives at a preset time interval, such as the position where a train arrives every 5-10 minutes, or each point on a newly developed track that requires position initialization. Objects with fixed positions may include common civil structures, equipment and facilities, and other targets with significant structural features and unchanged positions, such as signal lights, electrical boxes, kilometer signs, and other devices.

[0050] Step S2: extracting target semantic features and pixel position information of each target semantic feature in the image through a machine learning algorithm based on the collected image.

[0051] Step S3, based on the relationship between the internal and external parameters of the laser radar and the on-board image acquisition device, and the collected point cloud data, determine the relative position information of the train corresponding to the pixel position information of each target semantic feature, so as to obtain the corresponding point semantic information.

[0052] Step S4, using the location information of the point in the global map and the corresponding semantic information of the point as location comparison information to initialize the train position.

[0053] The train position initialization method provided in this embodiment uses an image acquisition device to obtain images and laser radar data collected when the train is located at the point selected in the line for train position initialization; based on the semantic features of the image target determined by the machine learning algorithm, such as common civil structures, equipment and facilities in the track area with significant structural features and unchanged positions, the laser radar data and the image acquisition device parameter fusion method are used to determine the relative position relationship between the pixel position information of each target and the train; finally, semantic information is constructed for the relative spatial positions of multiple targets, and is associated with the positions of the corresponding points in the global map, thereby realizing the train position initialization. The method provided in this embodiment uses the images and laser radar point clouds collected by the train image acquisition device as information sources, and does not require the construction of infrastructure such as transponders of traditional rail transit signal systems. Position comparison information is obtained by combining the position information of each preset point in the global map, thereby effectively initializing the train position, so that trains with different signal systems can be initialized on the same line, and trains with poor or even no signals can be initialized, effectively ensuring the availability and safety of train operation.

[0054] In this embodiment, the machine learning algorithm may be a target detection or instance segmentation algorithm, which is used to segment target features in an image and determine corresponding semantics. The target features may be targets with significant structural features and unchanged positions, such as common civil structures or equipment facilities in rail areas, such as tunnel entrances, kilometer markers, electrical boxes, signal machines, etc. The detection can obtain pixel position information of the semantic features in the image.

[0055] In one embodiment, a machine learning algorithm may be trained using images with artificially specified target type labels, such as CNN, YOLO, and other target detection algorithm networks. During the target detection process, the image pixel position of the corresponding target is obtained to determine the corresponding target type information.

[0056] In this embodiment, based on step S2, step S3 specifically includes the following steps:

[0057] Step S31, projecting the point cloud onto the image according to the pre-calibrated sensor internal and external parameter relationship between the laser radar and the image acquisition device, and obtaining the pixel position information of each semantic feature detected in the three-dimensional space;

[0058] Step S32: Taking the central pixel position of one of the semantic features as a reference position, relative pixel positions of other semantic features are determined.

[0059] In this embodiment, step S4 specifically includes the following steps:

[0060] Step S41, constructing a semantic-position dictionary based on the position information of the point in the global map and the corresponding semantic information of the point.

[0061] Step S42, when the train needs to implement position initialization, the surrounding images and point cloud data of the current position are collected by the on-board image acquisition device and the lidar respectively, and the image target semantic features and the relative pixel position information of the train corresponding to each target semantic feature are determined through steps S2 and S3, thereby determining the semantic information corresponding to the train position.

[0062] Step S43, compare and query the obtained train position semantic information with the semantic-position dictionary to determine the position information and use it as the train initialization position.

[0063] In this embodiment, a semantic-position dictionary is constructed using point semantic information and corresponding point position information as data pairs. During the position query process, the semantic information of the corresponding position determined by the train surrounding graphics collected by the train and the point cloud data is queried in the semantic-position dictionary to determine the corresponding position information, thereby realizing the determination of the train position. This method can quickly realize semantic query and determine the position information to ensure the safe operation of the train.

[0064] In this embodiment, in step S43, the semantic information and the semantic-position dictionary comparison query can be implemented by a similarity algorithm, and the similarity algorithm can specifically be a binary tree, Euclidean distance, cosine similarity, etc.

[0065] In one embodiment, the train position initialization process is explained by taking the signals of the entire line as initialization points, as follows.

[0066] In this embodiment, considering that the train may start and perform position initialization operations at different points on the line, initialization points can be set at certain intervals on the line, so that the train can find a point nearby that can be used for initialization without running a long distance, and then obtain its own position. In this embodiment, landmarks with significant features such as traffic lights are used as initialization points, such as Figure 2 As shown in the figure, the signal machine is a target that can be easily detected by the computer vision system and can divide the rail transit section. Therefore, each signal machine on the entire line is used as a point where initialization can be implemented.

[0067] Step A1: Use the signal lights of the entire line as initialization points, and collect surrounding images and point cloud data of the current position through the on-board camera and laser radar at the initialization points.

[0068] Specifically, the train runs in front of each signal and uses on-board cameras and lidar sensors to collect images and point cloud data of the environment near the signal.

[0069] Step A2: For camera images, use algorithms such as target detection or instance segmentation to obtain the image pixel positions of some artificially specified target types, thereby obtaining target semantic features. The artificially specified target types are common civil structures or equipment facilities in the rail area, such as tunnel entrances, kilometer markers, electrical boxes, signal machines, etc. Detection can obtain pixel position information of semantic features in the image, such as Figure 3 shown.

[0070] Step A3: Based on the previously calibrated internal and external parameter relationship between the laser radar and the camera, the point cloud data is projected onto the camera image to obtain the three-dimensional spatial pixel position information of the detected target semantic features, thereby determining the relative pixel position information of the train corresponding to each target semantic feature, and thus determining the semantic information of each point.

[0071] In a specific embodiment, if Figure 4 As shown in , if the central pixel position of one of the target semantic features is used as the reference position, the relative pixel positions of other target semantic features can be obtained, thereby constructing the semantic dictionary of the initialization point at the position. Figure 4 As shown in , the semantic dictionary of this point is: right-curve tunnel entrance, kilometer sign, electrical box, high-post dual-display signal. If the midpoint pixel of the high-post signal is used as the reference position (0,0,0), the pixel position information of the center point of other semantic features is: right-curve tunnel entrance relative pixel position coordinates (-2,10,1.5), kilometer sign relative pixel position coordinates (0,5,0), electrical box relative pixel position coordinates (0,0,-1). Similarly, the above operation is performed on all points used for position initialization in the line to obtain a dictionary that can query the position from the semantics.

[0072] Step A4: When the train needs to implement the position initialization function, the train uses the on-board camera and lidar to capture the image and point cloud of the scene in front, and adopts the same strategy as steps A2 and A3 to obtain the image semantic feature information, and then searches in the semantic-position dictionary, and obtains the position with the highest similarity as the train initialization position.

[0073] In another embodiment, the train position initialization process is explained by taking the stop signs of the entire line as initialization points, as follows.

[0074] Considering that the train may start and perform position initialization operations at different points on the line, initialization points can be set at certain intervals on the line so that the train can find a point nearby that can be used for initialization without running a long distance, and then obtain its own position. In this implementation, the stop signs at each station are used as initialization points, such as Figure 5As shown in the figure, the stop sign is a target that is easily detected by the computer vision system and can be used to divide the rail transit section and can be used as an initialization point.

[0075] Step B1: Use the signal lights of the entire line as initialization points, and collect surrounding images and point cloud data of the current position through the on-board camera and laser radar at the initialization points.

[0076] Specifically, the train runs to the front of the stop sign at each station and uses the on-board camera and lidar sensor to collect images and point cloud data of the environment near the stop sign.

[0077] Step B2: For camera images, use algorithms such as target detection or instance segmentation to obtain the image pixel positions of some artificially specified target types, thereby obtaining target semantic features. The artificially specified target types are common civil structures or equipment facilities in the rail area, such as tunnel entrances, kilometer markers, electrical boxes, signal machines, etc. Detection can obtain pixel position information of semantic features in the image, such as Figure 3 shown.

[0078] Step B3: Based on the previously calibrated internal and external parameter relationship between the laser radar and the camera, the point cloud data is projected onto the camera image to obtain the three-dimensional spatial pixel position information of the detected target semantic features, thereby determining the relative pixel position information of each target semantic feature corresponding to the train and determining the semantic information of each point.

[0079] In a specific embodiment, if Figure 6 As shown in , if the central pixel position of one of the target semantic features is used as the reference position, the relative positions of other target semantic features can be obtained, thereby constructing the semantic dictionary of the initialization point at the position. Figure 7 As shown in , the semantic dictionary of this point is: electrical box, high pole double display signal, stop sign. If the midpoint pixel of the stop sign is used as the reference position (0,0,0), the pixel position information of the center point of other semantic features is: electrical box relative pixel position coordinates (-3,1,-2), high pole double display signal relative pixel position coordinates (0,1,-1). Similarly, the above operation is performed on all points used for position initialization in the line to obtain a semantic-position dictionary that can query the position from the semantics.

[0080] Step B4: When the train needs to implement the position initialization function, the train uses the on-board camera and lidar to capture the image and point cloud of the scene in front, and adopts the same strategy as steps B2 and B3 to obtain the image semantic feature information, and then searches in the semantic-position dictionary, and obtains the position with the highest similarity as the train initialization position.

[0081] In order to solve the problem of position initialization of urban rail transit trains, this embodiment proposes a method of using computer vision to identify target semantic information in the train running line environment, constructing a "semantic-position" dictionary, and inferring the position of the current train in the global map by querying the dictionary, thereby achieving position initialization.

[0082] The advantages of this method are:

[0083] (1) There is no need to build infrastructure such as transponders of traditional rail transit signal systems.

[0084] (2) It allows trains with different signal systems to achieve position initialization on the same line.

[0085] (3) It allows trains without signal systems to achieve position initialization.

[0086] Device Embodiment

[0087] According to an embodiment of the present invention, a train position initialization device is provided, such as Figure 7 As shown, it is a block diagram of a train position initialization device provided in this embodiment. The train position initialization device according to an embodiment of the present invention includes:

[0088] The data acquisition module 10 is used to obtain images and point cloud data around multiple points preset in the track area through the on-board image acquisition device and the lidar sensor when the train is located at each point, wherein the image includes multiple objects with fixed positions.

[0089] In this embodiment, the point position may be a target position in a track line selected according to demand, and generally a landmark with significant features is used as the initialization point position, such as a signal light, a stop sign, or the position corresponding to other objects with fixed positions, or the position where a train arrives at a preset time interval, such as the position where a train arrives every 5-10 minutes, or each point on a newly developed track that requires position initialization. Objects with fixed positions may include common civil structures, equipment and facilities, and other targets with significant structural features and unchanged positions, such as signal lights, electrical boxes, kilometer signs, and other devices.

[0090] The feature extraction module 20 is used to extract the target semantic features and the pixel position information of each semantic feature in the image through a machine learning algorithm based on the collected image.

[0091] The semantic information determination module 30 is used to determine the relative position information of the train corresponding to the pixel position information of each target semantic feature based on the relationship between the internal and external parameters of the sensors of the laser radar and the vehicle-mounted image acquisition device, and the collected point cloud data, so as to obtain point semantic information.

[0092] The position initialization module 40 is used to realize train position initialization by using the position information of the point in the global map and the corresponding semantic information of the point as position comparison information.

[0093] In the train position initialization device provided in the present embodiment, the data acquisition module 10 uses an image acquisition device to obtain images and laser radar data collected when the train is located at the point selected in the line for train position initialization; the feature extraction module 20 and the semantic information determination module 30 determine the semantic features of the image target based on the machine learning algorithm, such as common civil structures, equipment and facilities in the track area, which have significant structural features and unchanged positions, and then use the laser radar data and camera parameter fusion method to determine the relative position relationship between the pixel position information of each target and the train; finally, semantic information is constructed for the relative spatial positions of multiple targets, and is associated with the positions of the corresponding points in the global map, so as to realize the train position initialization. In the present embodiment, images and laser radar point clouds collected by the train image acquisition device are used as information sources, and there is no need to build infrastructure such as transponders of traditional rail transit signal systems. Position comparison information is obtained by combining the position information of each preset point in the global map, so as to effectively initialize the train position, so that trains with different signal systems can be initialized on the same line, and trains with poor or even no signals can be initialized, which effectively ensures the availability and safety of train operation.

[0094] In this embodiment, in this embodiment, the machine learning algorithm can be a target detection or instance segmentation algorithm, which is used to segment target features in the image and determine the corresponding semantics, wherein the target features can be targets with significant structural features and unchanged positions, such as common civil structures or equipment facilities in the rail area, such as tunnel entrances, kilometer markers, electrical boxes, signal machines, etc. The detection can obtain the pixel position information of the semantic features in the image.

[0095] In one embodiment, a machine learning algorithm may be trained using images with artificially specified target type labels, such as CNN, YOLO, and other target detection algorithm networks. During the target detection process, the image pixel position of the corresponding target is obtained to determine the corresponding target type information.

[0096] In this embodiment, the semantic information determination module 30 is configured to perform the following steps to obtain point semantic information. The specific steps are as follows:

[0097] Step S31, projecting the point cloud onto the image according to the pre-calibrated sensor internal and external parameter relationship between the laser radar and the image acquisition device, and obtaining the pixel position information of each semantic feature detected in the three-dimensional space;

[0098] Step S32: Taking the central pixel position of one of the semantic features as a reference position, relative pixel positions of other semantic features are determined.

[0099] In this embodiment, the position initialization module 40 is configured to execute the following steps to realize the train initialization position, including the steps of:

[0100] Step S41, constructing a semantic-position dictionary based on the position information of the point in the global map and the corresponding semantic information of the point.

[0101] Step S42, when the train needs to implement position initialization, the surrounding images and point cloud data of the current position are collected by the on-board image acquisition device and the lidar respectively, and the image target semantic features and the relative pixel position information of the train corresponding to each target semantic feature are determined through steps S2 and S3, thereby determining the semantic information corresponding to the train position.

[0102] Step S43, compare and query the obtained train position semantic information with the semantic-position dictionary to determine the position information and use it as the train initialization position.

[0103] In this embodiment, a semantic-position dictionary is constructed using point semantic information and corresponding point position information as data pairs. During the position query process, the semantic information of the corresponding position determined by the train surrounding graphics collected by the train and the point cloud data is queried in the semantic-position dictionary to determine the corresponding position information, thereby realizing the determination of the train position. This method can quickly realize semantic query and determine the position information to ensure the safe operation of the train.

[0104] In this embodiment, in step S43, the semantic information and the semantic-position dictionary comparison query can be implemented by a similarity algorithm, and the similarity algorithm can specifically be a binary tree, Euclidean distance, cosine similarity, etc.

[0105] The embodiment of the present invention is an apparatus embodiment corresponding to the above-mentioned method embodiment. The specific operations of the processing steps of each module can be understood by referring to the description of the method embodiment, which will not be repeated here.

[0106] like Figure 8 As shown, the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the train position initialization method in the above embodiment is implemented.

[0107] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the train position initialization method in the above embodiment is implemented. Alternatively, when the computer program is executed by a processor, the train position initialization method in the above embodiment is implemented. When the computer program is executed by the processor, the following method steps are implemented:

[0108] Step S1, according to multiple points preset in the track area, when the train is located at each point, the surrounding image and point cloud data of each point are obtained through the on-board image acquisition device and the lidar sensor, wherein the image includes multiple objects with fixed positions.

[0109] Step S2: extracting target semantic features and pixel position information of each target semantic feature in the image through a machine learning algorithm based on the collected image.

[0110] Step S3, based on the relationship between the internal and external parameters of the laser radar and the on-board image acquisition device, and the collected point cloud data, determine the relative position information of the train corresponding to the pixel position information of each target semantic feature, so as to obtain the corresponding point semantic information.

[0111] Step S4, using the location information of the point in the global map and the corresponding semantic information of the point as location comparison information to initialize the train position.

[0112] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0113] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and the contents not described in detail in the specification of the present invention belong to the common knowledge of those skilled in the art.

Claims

1. A train position initialization method, characterized in that: The following steps are involved: According to multiple preset points in the track area, when the train is located at each point, the image and point cloud data around the point are obtained through the on-board image acquisition device and the laser radar sensor; Based on the collected images, the target semantic features and the pixel position information of each target semantic feature in the image are extracted by machine learning algorithm; According to the relationship between the internal and external parameters of the laser radar and the on-board image acquisition device, and the collected point cloud data, the pixel position information of each target semantic feature corresponds to the relative pixel position information of the train, thereby obtaining the corresponding point semantic information; and The train position is initialized using the location information of the point in the global map and the corresponding semantic information of the point as position comparison information.

2. The train position initialization method according to claim 1, characterized in that: Furthermore, according to the relationship between the internal and external parameters of the laser radar and the on-board image acquisition device, and the collected point cloud data, the pixel position information of each semantic feature corresponding to the relative pixel position information of the train is determined, so as to obtain the semantic information of each point, which specifically includes the following steps: According to the pre-calibrated relationship between the internal and external parameters of the laser radar and the image acquisition device, the point cloud is projected onto the image to obtain the pixel position information of each semantic feature detected in the three-dimensional space; The central pixel position of one of the semantic features is used as the reference position to determine the relative pixel positions of other semantic features.

3. The train position initialization method according to claim 1, characterized in that: The predetermined points are locations of landmarks with distinctive features; or The train arrives at the position of the ground track at preset time intervals; The acquired point peripheral image includes objects with fixed positions.

4. The train position initialization method according to claim 1, characterized in that: Further, the train position is initialized using the position information of the point in the global map and the corresponding semantic information of the point as position comparison information, which specifically includes the following steps: Construct a semantic-location dictionary based on the location information of the point in the global map and the corresponding semantic information of the point; When the train needs to implement position initialization, the surrounding images and point cloud data of the current position are collected by the on-board image acquisition device and the laser radar respectively, and the image target semantic features and the relative pixel position information of each target semantic feature corresponding to the train are determined, thereby determining the semantic information of the corresponding train position; as well as The obtained train position semantic information is compared with the semantic-position dictionary to determine the position information and use it as the train initialization position.

5. A train position initialization device, characterized in that: include: The data acquisition module is used to obtain images and point cloud data around the points through the on-board image acquisition device and the laser radar sensor according to multiple points preset in the track area when the train is located at each point; A feature extraction module is used to extract target semantic features and pixel position information of each semantic feature in the image through a machine learning algorithm based on the collected image; The semantic information determination module is used to determine the pixel position information of each target semantic feature corresponding to the relative position information of the train according to the relationship between the internal and external parameters of the sensors of the laser radar and the vehicle-mounted image acquisition device, and the collected point cloud data, so as to obtain the point semantic information; as well as The position initialization module is used to initialize the train position using the position information of the point in the global map and the corresponding point semantic information as position comparison information.

6. The train position initialization device according to claim 5, characterized in that: Further, the semantic information determination module is configured to execute the following steps to obtain the point semantic information, and the specific steps are as follows: According to the pre-calibrated relationship between the internal and external parameters of the laser radar and the image acquisition device, the point cloud is projected onto the image to obtain the pixel position information of each semantic feature detected in the three-dimensional space; The central pixel position of one of the semantic features is used as the reference position to determine the relative pixel positions of other semantic features.

7. The train position initialization device according to claim 5, characterized in that: The predetermined points are locations of landmarks with distinctive features; or The train arrives at the position of the ground track at preset time intervals; The acquired point peripheral image includes objects with fixed positions.

8. The train position initialization device according to claim 5, characterized in that: The position initialization module is further configured to perform the following steps to initialize the train position, including: Construct a semantic-location dictionary based on the location information of the point in the global map and the corresponding semantic information of the point; When the train needs to implement position initialization, the surrounding images and point cloud data of the current position are collected by the on-board image acquisition device and the laser radar respectively, the image target semantic features and the relative position information of the train corresponding to each target semantic feature are determined, and the semantic information corresponding to the train position is determined; and The obtained semantic information is compared with the semantic-position dictionary to obtain the position information as the train initialization position.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the train position initialization method as described in any one of claims 1 to 4 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the train position initialization method as described in any one of claims 1 to 4 is implemented.

Images