Obstacle detection method, system and equipment suitable for low-speed railway and medium

Abstract

The invention relates to an obstacle detection method, system, equipment and medium suitable for a low-speed railway, and belongs to the technical field of rail transit, and the method comprises the following steps: collecting image data and three-dimensional point cloud data of a rail in front of a train through a laser radar and a short-focus camera, and preprocessing the image data to obtain a pre-processed image; the preprocessed image data are transmitted into the trained rail-mounted area segmentation model, and a detected rail-mounted area result is output; mapping the detected image pixel points of the rail-mounted area to three-dimensional point cloud data to obtain a rail-mounted area point cloud; performing preprocessing and point cloud clustering on the point cloud data to detect non-specific obstacles; the three-dimensional space of the rail-mounted area is divided based on the point cloud of the rail-mounted area, foreign matter invasion judgment is conducted on the detected non-specific obstacle point cloud, if invasion occurs, obstacle and alarm information is sent to an automatic train protection system, autonomous obstacle detection of the train is achieved, and therefore the purpose of guaranteeing the operation safety of the train is achieved.

Description

Technical Field

[0001] The present disclosure belongs to the field of rail transit technology, and in particular relates to a method, system, device and medium suitable for detecting obstacles on low-speed railways. Background Art

[0002] my country's railway lines cover a wide area and have large horizontal and vertical spans. The train driving environment is complex and changeable. The intrusion of pedestrians, animals, fallen rocks, tools left behind by construction workers, foreign objects in the contact network, etc., pose a serious safety threat to running trains. The intrusion of foreign objects is random and difficult to detect. Therefore, obstacle detection in the track area is of great significance to ensure the safe operation of trains.

[0003] Currently, obstacles in the track area are detected by a single visual sensor, which has the following disadvantages:

[0004] 1. Low recognition reliability in bad weather or dim light;

[0005] 2. The targets that can be identified only include the trained target types. Untrained intrusion target types cannot be identified, and the universality is poor, making it difficult to fully ensure the safety of train operation;

[0006] 3. The amount of calculation is large, the hardware computing power is high, the recognition time is long, and the timeliness is low;

[0007] 4. The single sensor approach has low reliability and cannot meet the safety and reliability requirements of rail transit.

[0008] Therefore, it is necessary to provide a new method, system, equipment and medium suitable for low-speed railway obstacle detection to solve the above technical problems. Summary of the invention

[0009] The purpose of the present disclosure is to provide a method, system, device and medium suitable for low-speed railway obstacle detection in order to solve the above-mentioned problems.

[0010] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0011] A method for detecting obstacles on a low-speed railway comprises the following steps:

[0012] Obtain image data and 3D point cloud data of the track ahead of the train;

[0013] Preprocessing the image data to obtain preprocessed image data;

[0014] Input the preprocessed image data into a pre-built track area segmentation model, output the track area in the image, and extract image pixels of the track area;

[0015] Mapping the image pixel points of the track area to the three-dimensional point cloud data to obtain a track area point cloud;

[0016] Preprocessing the track area point cloud to obtain a preprocessed track area point cloud;

[0017] Performing clustering processing on the pre-processed track area point cloud to detect non-specific obstacles and obtain a non-specific obstacle point cloud;

[0018] Foreign object intrusion judgment is performed based on the non-specific obstacle point cloud and the track area point cloud to achieve obstacle detection.

[0019] As a further optimization solution of the present disclosure, image data and three-dimensional point cloud data of the track in front of the train are collected by a short-focus camera and a lidar installed on the train.

[0020] As a further optimization solution of the present disclosure, the image data is preprocessed to obtain the preprocessed image data including:

[0021] The image data is normalized and noise removed, and the size of the image is scaled to a size specified by the track area segmentation model to obtain pre-processed image data.

[0022] As a further optimization scheme of the present disclosure, the pre-processed image data is input into a pre-built track area segmentation model, the track area in the image is output, and the image pixel points of the track area are extracted, including:

[0023] The preprocessed image data is input into the pre-built and trained improved YOLOv8n-LD-seg track area segmentation model, the track area in the identified image is output, the bounding box of the track area is marked on the image, and the image pixel points R(u, v) of the track area are extracted.

[0024] As a further optimization scheme of the present invention, the improved YOLOv8n-LD-seg track area segmentation model uses YOLOv8n-seg as the baseline model, uses the DWR module in the ultra-real-time semantic segmentation model DWR-Seg in the backbone network to replace the Bottleneck module in the C2f module, and adds a large separable kernel attention LSKA between the maximum pooling layer and the second convolutional layer of the SPPF module to generate an improved YOLOv8-LD-seg track area segmentation model.

[0025] As a further optimization solution of the present disclosure, the image pixel points of the track area are mapped to the three-dimensional point cloud data to obtain the track area point cloud, which includes:

[0026] The image pixel points R(u, v) of the track area are mapped to the three-dimensional point cloud data to obtain the track area point cloud R(x, y, z); firstly, the internal parameter calibration of the short-focus camera is performed, and the parameters of the short-focus camera are determined using the Zhang Zhengyou chessboard calibration method; since the relative posture and displacement of the laser radar and the short-focus camera are fixed, the coordinate transformation relationship between the short-focus camera and the laser radar is established, and the 2D image data captured by the short-focus camera is represented by (U, V), and the 3D point cloud captured by the laser radar is represented by (X, Y, Z). By establishing a transformation matrix M, the 2D point (u, v) is mapped to the 3D point (x, y, z):

[0027]

[0028] Among them, the matrix (f u , f v ,u 0 , v 0 ) is the camera parameter, f u 、f v is the XY axis scale factor, u 0 , v 0 is the center point of the image plane, R is the rotation matrix, and t is the translation vector.

[0029] As a further optimization scheme of the present disclosure, preprocessing of the track area point cloud includes data superposition, point cloud downsampling, noise filtering and ground point filtering.

[0030] As a further optimization scheme of the present disclosure, clustering processing is performed on the pre-processed track area point cloud to detect non-specific obstacles, and the non-specific obstacle point cloud obtained includes:

[0031] The pre-processed track area point cloud is subjected to Euclidean clustering to detect non-specific obstacles:

[0032] Set the Euclidean distance threshold n, the minimum number of cluster points m, and the number of adjacent points k;

[0033] Randomly select point cloud p as seed point to join the set cluster C i , use the KD tree to find k adjacent points of point cloud p, calculate the Euclidean distance to p, and add points with Euclidean distance less than n to cluster C i , cluster C with cluster points greater than or equal to m i That is the detected non-specific obstacle point cloud.

[0034] 10. As a further optimization solution of the present disclosure, judging foreign body intrusion based on the non-specific obstacle point cloud and the track area point cloud includes:

[0035] For non-specific obstacle point cloud C i(x, y, z) and the track area point cloud R(x, y, z) are used to judge the intrusion of foreign objects: a three-dimensional track space R with height z fixed at [-3, 6] is established based on the track area point cloud. 3D , assuming that the x and y coordinates of the track area form a two-dimensional plane R 2D , and enclose it using the convex hull algorithm. The formula is as follows:

[0036] R 2D ={(x,y)|(x,y,z)∈R};

[0037] H={(x,y)|(x,y)∈ConvexHull(R 2D )};

[0038] Extend the two-dimensional boundary to the three-dimensional trajectory space R 3D , and fix the height range to [-3,6], the formula is as follows:

[0039] R 3D =(x,y,z)|(x,y)∈H,z∈[-3,6]};

[0040] Determine the non-specific obstacle point cloud C i Whether (x, y, z) is in the three-dimensional track space R of the track area 3D If C i (x,y,z)∈R 3D , an alarm message and obstacle detection results in the track area are issued.

[0041] An obstacle detection system for low-speed railways, comprising:

[0042] A data acquisition module, used to acquire image data and three-dimensional point cloud data of the track ahead of the train;

[0043] An image data preprocessing module, used to preprocess the image data to obtain preprocessed image data;

[0044] A track area segmentation module, used for inputting the pre-processed image data into a pre-built track area segmentation model, outputting the track area in the image, and extracting image pixels of the track area;

[0045] A mapping module, used to map the image pixels of the track area to the three-dimensional point cloud data to obtain a track area point cloud;

[0046] A point cloud preprocessing module, used for preprocessing the track area point cloud to obtain a preprocessed track area point cloud;

[0047] A clustering module, used for performing clustering processing on the pre-processed track area point cloud to detect non-specific obstacles and obtain a non-specific obstacle point cloud;

[0048] The foreign object intrusion judgment module is used to perform foreign object intrusion judgment based on the non-specific obstacle point cloud and the track area point cloud to achieve obstacle detection.

[0049] An electronic device comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0050] Memory, for storing computer programs;

[0051] The processor is used to execute the program stored in the memory to implement the obstacle detection method applicable to low-speed railways.

[0052] A computer-readable storage medium stores a computer program, which, when executed by a processor, is applicable to a low-speed railway obstacle detection method.

[0053] The beneficial effects of the present disclosure are:

[0054] The present disclosure uses image track areas to divide the original laser point cloud, simplifies the extraction of laser point cloud track areas, and avoids false alarms of obstacles in non-track areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 is a flow chart of a method in an embodiment of the present disclosure;

[0056] Figure 2 is a system structure block diagram in an embodiment of the present disclosure;

[0057] Figure 3 It is a block diagram of the device structure in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0058] The present application is further described in detail below in conjunction with the accompanying drawings. It is necessary to point out here that the following specific implementation methods are only used to further illustrate the present application and cannot be understood as limiting the scope of protection of the present application. Technical personnel in this field can make some non-essential improvements and adjustments to the present application based on the above application content.

[0059] like Figure 1 As shown, a method for detecting obstacles on a low-speed railway includes the following steps:

[0060] S1. Obtain image data and 3D point cloud data of the track ahead of the train:

[0061] The short-focus camera and lidar installed in the train driver's cab collect image data and 3D point cloud data of the track ahead of the train;

[0062] S2. Preprocess the image data to obtain preprocessed image data:

[0063] Preprocessing the image data, including normalization, noise removal, and scaling the image size to a size of 640*640 specified by the track area segmentation model;

[0064] S3, inputting the pre-processed image data into a pre-built track area segmentation model, outputting the track area in the image, and extracting image pixels of the track area:

[0065] The preprocessed image data is passed into the trained improved YOLOv8n-LD-seg track area segmentation model, the track area in the output recognition image is identified, the bounding box of the track area is marked on the image, and the image pixel points R(u,v) of the track area are extracted;

[0066] The improved YOLOv8n-LD-seg track area segmentation model uses YOLOv8n-seg as the baseline model, uses the DWR module in the ultra-real-time semantic segmentation model DWR-Seg in the backbone network to replace the Bottleneck module in the C2f module, and adds a large separable kernel attention LSKA between the maximum pooling layer and the second convolutional layer of the SPPF module to generate an improved YOLOv8-LD-seg track area segmentation model.

[0067] In the backbone network, the DWR module in the ultra-real-time semantic segmentation model DWR-Seg is used to replace the Bottleneck module in the C2f module. Specifically, the DWR module first inputs the features into a 3*3 convolution, and then uses the hole depth convolution with different hole rates to perform morphological filtering on regional features of different sizes through the branch structure, and then performs feature fusion through 1*1 point-by-point convolution to strengthen the connection between contextual features. The calculation formula is:

[0068] C 1 (x) = SiLU(BN(Conv(x)));

[0069] C 2 (x, d) = D d DConv(C 1 (x));

[0070] DWR(x)=PConv(BN(Γ d {C 2 (x, d)}))+x;

[0071] Among them, x is the input feature, Conv is a 3*3 convolution, BN is a Batch Normalization layer, SiLU is an activation function, and D d DConv has a 3*3 convolution in the depth direction with a dilation rate of d, PConv is a point-by-point convolution, Γ d is the concatenation operation of all d.

[0072] In the backbone network, a large separable kernel attention LSKA is added between the maximum pooling layer and the second convolution layer of the SPPF module. Specifically, the feature information of different scales is compressed and reduced in dimension through the pooling layer of the SPPF module, and then the feature values ​​in the local area are merged and passed to the LSKA module through the connection layer. LSKA decomposes the two-dimensional convolution kernel of the deep convolution layer into cascaded horizontal and vertical one-dimensional convolution kernels and spatial dilation convolution to generate the weighted features of the attention map. The output formula of LSKA is:

[0073]

[0074]

[0075] Among them, * and is the convolution and Hadamard product, d is the dilation rate; k represents the kernel size, is the input feature map, is the output result of the deep convolution. A is the output result of the expanded depth convolution. C is the point-by-point convolution of the global spatial information of the attention map, Γ d is the concatenation operation of all d.

[0076] S4, mapping the image pixels of the track area to the three-dimensional point cloud data to obtain the track area point cloud:

[0077] Map the detected image pixel points R(u, v) of the track area to the three-dimensional point cloud data to obtain the track area point cloud R(x, y, z);

[0078] The mapping process is as follows: first, the camera is calibrated internally, and the camera parameters are determined using Zhang Zhengyou's chessboard calibration method; secondly, since the relative posture and displacement of the radar and short-focus camera are fixed, the coordinate transformation relationship between the short-focus camera and the lidar is established. The image data captured by the short-focus camera is represented by (U, V), and the 3D point cloud captured by the lidar is represented by (X, Y, Z). By establishing a transformation matrix M, the 2D point (u, v) is mapped to the 3D point (x, y, z):

[0079]

[0080] In the formula, the matrix (fu , f v ,u 0 , v 0 ) is the camera parameter, f u 、f v is the XY scale factor (effective focal length in horizontal and vertical directions), u 0 , v 0 is the center point of the image plane. R is the rotation matrix and t is the translation vector.

[0081] S5. Preprocess the track area point cloud to obtain a preprocessed track area point cloud:

[0082] Data overlay (i.e., to enhance the stability and integrity of point cloud data, overlay the current frame with the previous 3 frames of data),

[0083] Point cloud downsampling (select voxel size (such as 0.1 meters), divide the entire point cloud space into many small voxels; for each point in the voxel, calculate its average coordinates, and use the average value to replace all the original points; form downsampled point cloud data to reduce the amount of point cloud data);

[0084] Outlier point filtering: for each point, the point and its 50 nearest points are combined into a point set; assuming that the point set conforms to the normal distribution, the mean and standard deviation of the point set are calculated; traverse the points in the point set, if the point is within 1 standard deviation, then keep the point; otherwise, remove it;

[0085] Ground point filtering (randomly select 3 points in the point cloud, and the 3 points form a plane. Use these points to calculate the plane's normal vector and equation (Ax+By+Cz+D=0), traverse all points in the point cloud, and calculate the distance from each point to the fitting plane. Mark the points whose distance is less than the preset threshold n as ground points. Repeat the above process, set the maximum number of iterations k, and when the number of iterations reaches k or the proportion of ground points reaches the threshold m, find A, B, C, D corresponding to the largest number of ground points as the parameters of the plane.)

[0086] S6. Clustering the pre-processed track area point cloud to detect non-specific obstacles and obtain a non-specific obstacle point cloud:

[0087] The preprocessed track area point cloud is clustered in Euclidean order to detect non-specific obstacles. First, the Euclidean distance threshold n, the minimum number of cluster points m, and the number of adjacent points k are set. Point cloud p is randomly selected as a seed point to be added to cluster C. i , use the KD tree to find k neighboring points of point p, calculate the Euclidean distance to p, and add points with Euclidean distance less than n to cluster C i , C with cluster points greater than or equal to m i These are non-specific obstacles that are detected.

[0088] S7. Perform foreign body intrusion judgment based on the non-specific obstacle point cloud and the track area point cloud to achieve obstacle detection:

[0089] Point cloud C of non-specific obstacles detected i (x, y, z) and the track area point cloud R(x, y, z) are used to judge the intrusion of foreign objects. First, a three-dimensional track space with height z fixed at [-3, 6] is established according to the track area point cloud data. It is assumed that the x and y coordinates of the track area form a two-dimensional plane, and the convex hull algorithm is used to enclose these points.

[0090] R 2D ={(x,y)|(x,y,z)∈R};

[0091] H={(x,y)|(x,y)∈ConvexHull(R 2D )};

[0092] Then, we extend this two-dimensional boundary into three-dimensional space and fix the height range to [-3, 6]:

[0093] R 3D =(x,y,z)|(x,y)∈H,z∈[-3,6]};

[0094] Determine obstacle point cloud C i (x, y, z) is in the three-dimensional space R of the track area 3D If C i (x,y,z)∈R 3D , an alarm message and the obstacle detection results in the track area will be sent to the train automatic protection system.

[0095] like Figure 2 As shown, an embodiment of the present disclosure provides an obstacle detection system applicable to a low-speed railway, comprising:

[0096] A data acquisition module, used to acquire image data and three-dimensional point cloud data of the track ahead of the train;

[0097] The data acquisition module consists of an on-board host, a short-focus camera, and a laser radar. The on-board host has a built-in Ethernet switching module, which communicates with the automatic train protection system via Ethernet in the form of a user datagram protocol. The data from the laser radar and the short-focus camera are received through the communication interface, and then integrated to detect obstacles in the track area. The short-focus camera is used to collect video image information of the train's forward running environment. The laser radar is used to collect three-dimensional information of the train's forward running environment.

[0098] An image data preprocessing module, used to preprocess the image data to obtain preprocessed image data;

[0099] A track area segmentation module, used for inputting the pre-processed image data into a pre-built track area segmentation model, outputting the track area in the image, and extracting image pixels of the track area;

[0100] A mapping module, used to map the image pixels of the track area to the three-dimensional point cloud data to obtain a track area point cloud;

[0101] A point cloud preprocessing module, used for preprocessing the track area point cloud to obtain a preprocessed track area point cloud;

[0102] A clustering module, used for performing clustering processing on the pre-processed track area point cloud to detect non-specific obstacles and obtain a non-specific obstacle point cloud;

[0103] The foreign object intrusion judgment module is used to perform foreign object intrusion judgment based on the non-specific obstacle point cloud and the track area point cloud to achieve obstacle detection.

[0104] The implementation process of the functions and effects of each module in the above system is specifically described in the implementation process of the corresponding steps in the above method, which will not be repeated here.

[0105] For the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The system embodiment described above is only schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0106] In the above-mentioned embodiment, any multiple of all modules can be combined in one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least some functions of one or more of these modules can be combined with at least some functions of other modules and implemented in one module. At least one of all modules can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation modes of software, hardware and firmware or in a suitable combination of any of them. Alternatively, at least one of all modules can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be executed.

[0107] See also Figure 3 , an electronic device provided by an embodiment of the present disclosure includes a processor 1110, a communication interface 1120, a memory 1130 and a communication bus 1140, wherein the processor 1110, the communication interface 1120, and the memory 1130 communicate with each other through the communication bus 1140;

[0108] Memory 1130, for storing computer programs;

[0109] The processor 1110 is used to implement the obstacle detection method applicable to low-speed railways as shown below when executing the program stored in the memory 1130.

[0110] The communication bus 1140 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus 1140 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0111] The communication interface 1120 is used for communication between the above electronic device and other devices.

[0112] The memory 1130 may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory 1130 may also be at least one storage device located away from the processor 1110.

[0113] The above-mentioned processor 1110 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0114] The embodiment of the present disclosure further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned obstacle detection method applicable to low-speed railways is implemented.

[0115] The computer-readable storage medium may be included in the device / apparatus described in the above embodiment; or it may exist independently without being assembled into the device / apparatus. The above computer-readable storage medium carries one or more programs. When the above one or more programs are executed, the obstacle detection method applicable to low-speed railway according to the embodiment of the present disclosure is implemented.

[0116] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: 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), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0117] The above-mentioned embodiments only express several implementation methods of the present disclosure, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present disclosure. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present disclosure, and these all belong to the protection scope of the present disclosure.

Claims

1. A method for detecting obstacles on low-speed railways, characterized in that: The following steps are involved: Obtain image data and 3D point cloud data of the track ahead of the train; Preprocessing the image data to obtain preprocessed image data; Input the preprocessed image data into a pre-built track area segmentation model, output the track area in the image, and extract image pixels of the track area; Mapping the image pixel points of the track area to the three-dimensional point cloud data to obtain a track area point cloud; Preprocessing the track area point cloud to obtain a preprocessed track area point cloud; Performing clustering processing on the pre-processed track area point cloud to detect non-specific obstacles and obtain a non-specific obstacle point cloud; Foreign object intrusion judgment is performed based on the non-specific obstacle point cloud and the track area point cloud to achieve obstacle detection.

2. The method for detecting obstacles on low-speed railways according to claim 1, characterized in that: The short-focus camera and lidar installed on the train collect image data and three-dimensional point cloud data of the track in front of the train.

3. The method for detecting obstacles on low-speed railways according to claim 1, characterized in that: Preprocessing the image data to obtain preprocessed image data includes: The image data is normalized and noise removed, and the size of the image is scaled to a size specified by the track area segmentation model to obtain pre-processed image data.

4. The method for detecting obstacles on low-speed railways according to claim 1, characterized in that: Inputting the preprocessed image data into a pre-built track area segmentation model, outputting the track area in the image, and extracting image pixels of the track area includes: The preprocessed image data is input into the pre-built and trained improved YOLOv8n-LD-seg track area segmentation model, the track area in the identified image is output, the bounding box of the track area is marked on the image, and the image pixel points R(u,v) of the track area are extracted.

5. The method for detecting obstacles on a low-speed railway according to claim 4, characterized in that: The improved YOLOv8n-LD-seg track area segmentation model uses YOLOv8n-seg as the baseline model, uses the DWR module in the ultra-real-time semantic segmentation model DWR-Seg in the backbone network to replace the Bottleneck module in the C2f module, and adds a large separable kernel attention LSKA between the maximum pooling layer and the second convolutional layer of the SPPF module to generate an improved YOLOv8-LD-seg track area segmentation model.

6. The obstacle detection method applicable to low-speed railway according to claim 1, characterized in that: Mapping the image pixel points of the track area to the three-dimensional point cloud data to obtain the track area point cloud includes: The image pixel points R(u,v) of the track area are mapped to the three-dimensional point cloud data to obtain the track area point cloud R(x,y,z); firstly, the internal parameter calibration of the short-focus camera is performed, and the parameters of the short-focus camera are determined using the Zhang Zhengyou chessboard calibration method; since the relative posture and displacement of the laser radar and the short-focus camera are fixed, the coordinate transformation relationship between the short-focus camera and the laser radar is established, and the 2D image data captured by the short-focus camera is represented by (U,V), and the 3D point cloud captured by the laser radar is represented by (X,Y,Z). By establishing a transformation matrix M, the 2D point (u,v) is mapped to the 3D point (x,y,z): Among them, the matrix (f u , f v , u0, v0) are the camera parameters, f u 、f v is the scale factor in the XY axis direction, u0, v0 are the center points of the image plane, R is the rotation matrix, and t is the translation vector.

7. The method for detecting obstacles on a low-speed railway according to claim 1, characterized in that: The preprocessing of the track area point cloud includes data superposition, point cloud downsampling, noise filtering and ground point filtering.

8. The obstacle detection method applicable to low-speed railway according to claim 1, characterized in that: The pre-processed track area point cloud is clustered to detect non-specific obstacles, and the obtained non-specific obstacle point cloud includes: The pre-processed track area point cloud is subjected to Euclidean clustering to detect non-specific obstacles: Set the Euclidean distance threshold n, the minimum number of cluster points m, and the number of adjacent points k; Randomly select point cloud p as seed point to join the set cluster C i , use KD tree to find k adjacent points of point cloud p, calculate the Euclidean distance with p, and add points with Euclidean distance less than n to cluster C i , cluster C with cluster points greater than or equal to m i That is the detected non-specific obstacle point cloud.

9. The method for detecting obstacles on a low-speed railway according to claim 1, characterized in that: The foreign body intrusion judgment based on the non-specific obstacle point cloud and the track area point cloud includes: For non-specific obstacle point cloud C i (x, y, z) and the track area point cloud R(x, y, z) are used to judge the intrusion of foreign objects: a three-dimensional track space R with height z fixed at [-3, 6] is established based on the track area point cloud. 3D , assuming that the x and y coordinates of the track area form a two-dimensional plane R 2D , and enclose it using the convex hull algorithm. The formula is as follows: R 2D ={(x,y)|(x,y,z)∈R}; H={(x,y)|(x,y)∈ConvexHull(R 2D )}; Extend the two-dimensional boundary to the three-dimensional trajectory space R 3D , and fix the height range to [-3,6], the formula is as follows: R 3D =(x,y,z)|(x,y)∈H,z∈[-3,6]}; Determine the non-specific obstacle point cloud C i Whether (x, y, z) is in the three-dimensional track space R of the track area 3D If C i (x,y,z)∈R 3D , an alarm message and obstacle detection results in the track area are issued.

10. An obstacle detection system suitable for low-speed railways, characterized in that: include: A data acquisition module, used to acquire image data and three-dimensional point cloud data of the track ahead of the train; An image data preprocessing module, used to preprocess the image data to obtain preprocessed image data; A track area segmentation module, used for inputting the pre-processed image data into a pre-built track area segmentation model, outputting the track area in the image, and extracting image pixels of the track area; A mapping module, used to map the image pixels of the track area to the three-dimensional point cloud data to obtain a track area point cloud; A point cloud preprocessing module, used for preprocessing the track area point cloud to obtain a preprocessed track area point cloud; A clustering module, used for performing clustering processing on the pre-processed track area point cloud to detect non-specific obstacles and obtain a non-specific obstacle point cloud; The foreign object intrusion judgment module is used to perform foreign object intrusion judgment based on the non-specific obstacle point cloud and the track area point cloud to achieve obstacle detection.

11. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, for storing computer programs; A processor is used to execute the program stored in the memory to implement the obstacle detection method applicable to low-speed railways described in any one of claims 1-9.

12. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for detecting obstacles applicable to low-speed railways described in any one of claims 1 to 9 is implemented.

Images