Method and device for determining foreign matter at bottom of carriage and carriage detection system

Abstract

The invention provides a compartment bottom foreign matter determination method and device and a compartment detection system. The method comprises the steps that a right-handed coordinate system of a compartment is established, the positive direction of the X axis of the right-handed coordinate system is parallel to the advancing direction of a train, the Z axis of the right-handed coordinate system is parallel to the vertical direction, and the Y axis of the right-handed coordinate system is perpendicular to the X axis and the Z axis; point cloud data obtained by scanning the carriage through the laser scanner is obtained, and the point cloud data is coordinate data of a right-handed coordinate system; determining whether target point cloud data of which the Z-axis coordinates are greater than or equal to a preset height threshold exists in the point cloud data or not; and under the condition of determining that the target point cloud data exists in the point cloud data, determining that the foreign matter exists at the bottom of the carriage. According to the scheme, the problem of low efficiency of train compartment bottom foreign matter detection in the prior art is solved.

Description

Technical Field

[0001] The present application relates to the field of vehicle measurement technology, and in particular to a method and device for determining foreign objects at the bottom of a vehicle compartment, a computer program product, and a vehicle compartment detection system. Background Art

[0002] At present, most coal carriages are monitored under the vehicle by installing high-speed cameras. Personnel in the undercarriage monitoring room monitor by watching and replaying videos in real time. Foreign objects under the vehicle are detected by human observation and timely feedback is given to the command center to achieve early warning. Due to the lack of intelligent detection technology, the quality of related detection work is overly dependent on the sense of responsibility, work attitude and carefulness of the detection personnel in this foreign object extraction based on undercarriage monitoring video. The monitoring work is intensive and the damage of foreign objects cannot be warned in real time. Missed reports and false reports often occur, which is time-consuming and labor-intensive. At the same time, when the train is traveling at high speed, the manual detection method requires high work intensity of the monitoring personnel and is affected by the mental state of the personnel, making it difficult to guarantee the quality and efficiency of related monitoring. Summary of the invention

[0003] The main purpose of the present application is to provide a method, device, computer program product and carriage detection system for determining foreign objects at the bottom of a carriage, so as to at least solve the problem of low efficiency of foreign object detection at the bottom of a train carriage in the prior art.

[0004] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for determining foreign objects at the bottom of a car is provided, comprising: establishing a right-handed coordinate system of the car, the positive direction of the X-axis of the right-handed coordinate system is parallel to the forward direction of the train, the Z-axis of the right-handed coordinate system is parallel to the vertical direction, and the Y-axis of the right-handed coordinate system is perpendicular to the X-axis and the Z-axis respectively; acquiring point cloud data obtained by scanning the car with a laser scanner, wherein the point cloud data is coordinate data of the right-handed coordinate system; determining whether there is target point cloud data in the Z-axis coordinate greater than or equal to a preset height threshold in the point cloud data; and determining that there is a foreign object at the bottom of the car when it is determined that the target point cloud data exists in the point cloud data.

[0005] Optionally, determining whether there is target point cloud data in the point cloud data whose Z-axis coordinate is greater than or equal to a preset height threshold includes: performing grayscale processing on the point cloud data to obtain a horizontal cross-sectional image; removing the point cloud data of the border of the car in the horizontal cross-sectional image to obtain a point cloud set; determining whether there is target point cloud data in the point cloud set whose Z-axis coordinate is greater than or equal to the preset height threshold to obtain a determination result.

[0006] Optionally, grayscale processing is performed on the point cloud data to obtain a horizontal cross-sectional image, including: intercepting a point cloud whose length is less than or equal to a preset length from the point cloud data to generate an initial horizontal cross-sectional image; extracting the maximum value of the X axis and the minimum value of the X axis in the initial horizontal cross-sectional image, and extracting the maximum value of the Y axis and the minimum value of the Y axis in the initial horizontal cross-sectional image; creating a blank image, wherein the resolution of the blank image is A×B, and the pixel value in the horizontal direction is A=INT((x max -x min ) / 0.01), the vertical pixel value is B=INT((y max -y min ) / 0.01), x max represents the maximum value of the X axis in the initial horizontal cross-sectional image, x min represents the minimum value of the X axis in the initial horizontal cross-sectional image, y max represents the maximum value of the Y axis in the initial horizontal cross-sectional image, y min represents the minimum value of the Y axis in the initial horizontal cross-sectional image; the point cloud coordinates in the initial horizontal cross-sectional image are transferred to the blank image with (x min ,y min ) is the target coordinate system with the origin; the point cloud coordinates in the target coordinate system are converted into image coordinates; the grayscale values ​​of the pixels in the image coordinates that have the point cloud data are adjusted to the first pixel value, and the grayscale values ​​of the pixels in the image coordinates that do not have the point cloud data are adjusted to the second pixel value, so as to obtain the horizontal cross-sectional image.

[0007] Optionally, the point cloud data of the border of the car in the horizontal section image is removed to obtain a point cloud set, including: cutting the horizontal section image to obtain multiple cut sets; removing the point cloud data of the border of the car in each of the cut sets to obtain the point cloud set.

[0008] Optionally, the horizontal section image is cropped to obtain multiple cropping sets, including: obtaining pixel values ​​of the point cloud data in the horizontal section image; obtaining a cutting position, wherein the cutting position is a position where the pixel values ​​of all the point cloud data in N consecutive columns are target pixel values, N≥1; and according to the cutting position, the horizontal section image is cropped to obtain multiple cropping sets.

[0009] Optionally, the point cloud data of the border of the carriage in each of the cut-off sets are removed to obtain the point cloud set, including: extracting the border point cloud data that meets preset conditions in the cut-off set, wherein the preset conditions include one or more of the following: the point cloud data is the maximum value of the X-axis, the point cloud data is the minimum value of the X-axis, the deviation of the normal vector of the point cloud data from the X-axis is less than or equal to a preset angle, the difference between the point cloud data and the maximum value of the X-axis is less than or equal to a preset difference, the difference between the point cloud data and the minimum value of the X-axis is less than or equal to the preset difference, the point cloud data is the maximum value of the Y-axis, the point cloud data is the minimum value of the Y-axis, the deviation of the normal vector of the point cloud data from the Y-axis is less than or equal to the preset angle, the difference between the point cloud data and the maximum value of the Y-axis is less than or equal to the preset difference, and the difference between the point cloud data and the minimum value of the Y-axis is less than or equal to the preset difference; deleting the border point cloud data that meets the preset conditions in all the cut-off sets to obtain the point cloud set.

[0010] Optionally, after determining that there is a foreign object at the bottom of the carriage, the method further includes: generating an alarm message when it is determined that there is a foreign object at the bottom of the carriage; and sending the alarm message to a target terminal.

[0011] According to another aspect of the present application, a device for determining foreign objects at the bottom of a carriage is provided, comprising: an establishing unit for establishing a right-handed coordinate system of the carriage, wherein the positive direction of the X-axis of the right-handed coordinate system is parallel to the forward direction of the train, the Z-axis of the right-handed coordinate system is parallel to the vertical direction, and the Y-axis of the right-handed coordinate system is perpendicular to the X-axis and the Z-axis respectively; an acquiring unit for acquiring point cloud data obtained by scanning the carriage with a laser scanner, wherein the point cloud data is coordinate data of the right-handed coordinate system; a first determining unit for determining whether there is target point cloud data in the Z-axis coordinate greater than or equal to a preset height threshold in the point cloud data; and a second determining unit for determining that there is a foreign object at the bottom of the carriage when it is determined that the target point cloud data exists in the point cloud data.

[0012] According to another aspect of the present application, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the steps of any one of the methods for determining a foreign object at the bottom of a vehicle compartment are implemented.

[0013] According to another aspect of the present application, a vehicle compartment detection system is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a method for executing any one of the foreign object determination methods at the bottom of the vehicle compartment.

[0014] By applying the technical solution of the present application, manual inspection is no longer required. By installing a laser scanner and scanning the car, point cloud data can be obtained. The places with relatively high heights in the point cloud data are where foreign objects are protruding. For example, if there is an extra water cup on the table, the water cup is a foreign object. This solution does not require complex machine learning and deep learning, nor does it require straight line fitting and plane fitting. It greatly simplifies the process of foreign object detection, improves the speed of foreign object detection, and thus improves the efficiency of foreign object detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings constituting part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0016] Figure 1 A hardware structure block diagram of a mobile terminal for executing a method for determining a foreign object at the bottom of a vehicle compartment provided in an embodiment of the present application is shown;

[0017] Figure 2 A schematic flow chart of a method for determining foreign matter at the bottom of a vehicle compartment provided according to an embodiment of the present application is shown;

[0018] Figure 3 A schematic diagram of the first right-hand system is shown;

[0019] Figure 4 A schematic diagram of a second right-hand system is shown;

[0020] Figure 5 A side view of the carriage is shown;

[0021] Figure 6 A top view of the carriage is shown;

[0022] Figure 7 A horizontal cross section of the carriage is shown;

[0023] Figure 8 A structural block diagram of a foreign object determination device at the bottom of a vehicle compartment provided according to an embodiment of the present application is shown.

[0024] The above drawings include the following reference numerals:

[0025] 102, processor; 104, memory; 106, transmission device; 108, input and output devices. DETAILED DESCRIPTION

[0026] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0027] In order to enable those skilled in the art to better understand the solution of the present application, 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 only 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 should fall within the scope of protection of the present application.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] As mentioned in the background technology, there is an urgent need to propose a point cloud-based coal carriage segmentation and foreign object extraction method. The automatic segmentation of carriage point clouds and the automatic extraction of carriage foreign objects are of great significance for improving the quality of monitoring work, increasing the monitoring of foreign objects under the carriage of empty return vehicles, and significantly improving the automation level of the monitoring system, thereby ensuring the quality of monitoring operations.

[0030] At present, point cloud segmentation methods mainly include traditional point cloud segmentation methods and deep learning-based segmentation methods. Traditional point cloud segmentation algorithms include: methods based on geometric features, methods based on region growing, etc.; deep learning-based point cloud segmentation methods include: PointNet point cloud segmentation algorithm, RSNet point cloud segmentation algorithm, PointCNN point cloud segmentation algorithm and KPConv point cloud segmentation algorithm, etc.

[0031] Traditional point cloud segmentation algorithm: The method based on geometric features mainly uses the geometric features of point clouds (such as normals, curvature, curvature changes, etc.) for segmentation. The normal direction of each point is calculated, and the angle between normals or normal changes are used to identify the segmentation boundary in the point cloud. The curvature can describe the local geometric features of the point cloud surface, and the different regions or objects in the point cloud are determined by detecting the curvature changes. The method based on region growing regards the point cloud as a collection of dense regions, and achieves segmentation by combining adjacent points into regions with similar attributes.

[0032] PointNet and improved point cloud segmentation algorithm: In 2017, a deep learning-based point cloud processing framework PointNet was proposed. It was the first time that a method was proposed to operate directly on point clouds instead of converting them into voxel grids or other representations. The local features of each point were processed through projection and aggregation operations, which has strong universality and flexibility. In the same year, PointNet++, an evolution of PointNet, was proposed. This method introduced a hierarchical feature learning structure, which improved the accuracy and efficiency of point cloud segmentation and scene understanding through step-by-step aggregation and resampling. It can process point cloud data of different scales and complexities and is suitable for various three-dimensional recognition tasks. Based on the PointNet method, the Frustum PointNet method was proposed in 2018, which is specially designed to detect objects from 2D images and segment and identify them in three-dimensional point clouds. It extracts object candidates by utilizing the frustum generated from the viewpoint, and then applies PointNet to semantically segment and recognize these candidates.

[0033] RSNet Point Cloud Segmentation Algorithm: RSNet, a powerful and conceptually concise network published in 2018, is designed for 3D point cloud segmentation tasks. In order to overcome the challenges brought by the sparsity and inhomogeneity of point cloud data, RSNet designs an effective mechanism for combining global and local features. This combination enables the network to focus on both the overall structure and local details of the point cloud, improving the semantic segmentation accuracy and generalization ability of point cloud data, with the goal of processing complex 3D point cloud data, especially those with multiple categories and irregular shapes.

[0034] PointCNN point cloud segmentation algorithm: The PointCNN point cloud segmentation algorithm was proposed in 2018. The algorithm aims to solve the problems of disorder and transformation invariance in point cloud data. It introduces a novel convolution operation that can perform feature extraction in non-Euclidean space (ie, point cloud space). This convolution operation takes into account the position and arrangement of points in the point cloud, rather than simply converting the point cloud into a regular structure for processing, thereby more effectively capturing the local and global features of the point cloud, and improving the representation ability of the point cloud by adaptively learning the neighborhood structure and features of each point. This mechanism enables PointCNN to adapt to point cloud data of different scales and complexities, thereby improving the accuracy and robustness of segmentation and recognition.

[0035] KPConv point cloud segmentation algorithm: In 2019, a new convolutional neural network structure based on point cloud was proposed, and the concept of Kernel Point was introduced. That is, in the convolution process, each point cloud point is regarded as a Kernel Point, and the relationship between these points is used to define the convolution kernel. This method can effectively handle the changes in the position, density and arrangement of points in the point cloud, thereby improving the flexibility and adaptability of the convolution operation. By dynamically adjusting the size and shape of the convolution kernel at each point, adaptive learning of the local structure of the point cloud is achieved. This enables the algorithm to capture local features of different scales and complexities, improves the algorithm's ability to understand the overall structure of the point cloud, and also helps solve the scale and shape changes in point cloud data.

[0036] According to the classification and summary of their technical and application characteristics, the current point cloud-based foreign body extraction methods are mainly divided into: geometric feature method, support vector machine method, density and statistical feature-based method, multi-sensor data combination method and segmentation-based method.

[0037] Methods based on geometric shape descriptors: Using FPFH (Fast Point Feature Histograms) or similar local feature descriptors to distinguish different objects or detect outliers is a common point cloud processing method. Local feature descriptors such as FPFH are used to describe the local geometric structure around each point in the point cloud. They calculate the feature vector of each point, which can capture the local shape information in the point cloud data, such as curvature, normal direction, etc. Outliers usually do not have similar FPFH features with surrounding points because they may represent unusual geometric structures or noise in the point cloud. Therefore, outliers can be identified and filtered out by comparison. This method is widely used in tasks such as object recognition, classification, and outlier detection in point cloud processing.

[0038] Traditional machine learning methods: The application of support vector machines (SVM) on point cloud data mainly involves using it as a classifier to identify and extract specific types of foreign objects or abnormal points. For supervised learning tasks, the point cloud data must be labeled with its category, such as normal points or abnormal points. Select an appropriate SVM kernel function and train the SVM model to learn the distinguishing boundary between normal points and abnormal points by taking the feature vector as input. Use the trained SVM model to make predictions on the point cloud data. For each point, the SVM outputs a category label or anomaly score to determine whether it is an abnormal point.

[0039] Methods based on density and statistical features: Use the density information of points to detect and identify foreign objects or abnormal points. This method is usually combined with clustering or density estimation algorithms. First, it is necessary to estimate the density of each point in the point cloud. Common density estimation methods include radius-based methods (such as the number of points within the radius) or neighborhood density-based methods (such as the number of K neighboring points). Based on the density information of each point in the point cloud, clustering algorithms (such as DBSCAN) or other density estimation methods (such as LOF) are used to identify points that are significantly different from the surrounding environment. This method requires the selection of appropriate radius parameters and neighbor number parameters to ensure good detection results, and the identified abnormal points usually require further post-processing steps.

[0040] Method of combining multi-sensor data: Combine visual information and spatial information, and match and register the pixels in the RGB image with the three-dimensional points in the point cloud through external or internal calibration between the camera and the lidar. Combine the color information of the RGB image with the spatial information of the point cloud (such as the position and density of the points). This step uses the visual information provided by the camera to enhance the semantic understanding and environmental perception capabilities of the point cloud data, and improve the accuracy and robustness of foreign body extraction. Use the visual features of the RGB image, such as color, texture, shape, etc., to extract and describe potential foreign body areas. Use the spatial information of the point cloud data, such as the density and normal vector of the points, combined with the visual features for comprehensive analysis, which helps to eliminate or confirm possible misdetected areas.

[0041] Segmentation-based methods: Pixel-level semantic segmentation is performed through deep learning models (such as neural networks), and features such as point position, normal vector, color, etc. are extracted from point cloud data to segment the point cloud into different semantic categories. Define the semantic categories of foreign objects or abnormal points that need to be identified, such as vehicles, pedestrians, buildings, etc. The segmented point cloud data is compared and classified with predefined categories, and the identified abnormal points or foreign objects are marked or extracted in the point cloud data.

[0042] Among the above methods, the method based on geometric shape descriptors has limitations and disadvantages in practical applications in terms of computational efficiency, parameter setting, scope of application, and dependence on data quality and structure; the traditional machine learning method has high computational complexity, especially in the case of high-dimensional feature space or large data volume, the time cost of training and prediction increases significantly, so it is not suitable for large-scale data; the method based on density and statistical features is more sensitive to the selection of parameters, and inappropriate parameter selection may lead to missed detection or false detection. It depends on density information, so the method may show different effects for different types of anomalies and foreign objects; in order to effectively fuse RGB images and point cloud data, the method combining multi-sensor data needs to ensure that the data quality and calibration of different sensors are consistent. How to effectively fuse this information requires the design of appropriate fusion strategies and algorithms; the foreign body extraction method based on point cloud segmentation will be affected by the presence of noise, occlusion or incompleteness in the point cloud data, and there are also problems such as insufficient processing of boundaries and details between different categories.

[0043] As introduced in the background technology, the efficiency of foreign object detection at the bottom of train compartments in the prior art is low. To solve the above problem, the embodiments of the present application provide a method, device, computer program product and compartment detection system for determining foreign objects at the bottom of a compartment.

[0044] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0045] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 1 is a hardware structure block diagram of a mobile terminal for determining a foreign object at the bottom of a vehicle compartment according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0046] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the display method of device information in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The transmission device 106 is used to receive or send data via a network. The above-mentioned specific examples of the network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0047] In this embodiment, a method for determining foreign objects at the bottom of a vehicle compartment that runs on a mobile terminal, a computer terminal, or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0048] Figure 2 FIG. 1 is a flow chart of a method for determining foreign matter at the bottom of a vehicle compartment according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0049] Step S201, establishing a right-handed coordinate system of the carriage, wherein the positive direction of the X-axis of the right-handed coordinate system is parallel to the forward direction of the train, the Z-axis of the right-handed coordinate system is parallel to the vertical direction, and the Y-axis of the right-handed coordinate system is perpendicular to the X-axis and the Z-axis respectively;

[0050] Specifically, according to the installation angle of the scanner, the train's forward direction is defined as the positive direction of the X-axis, the zenith direction is defined as the positive direction of the Z-axis, and the X-axis, Y-axis, and Z-axis form a right-handed system. In the three-dimensional coordinate system, the right-handed system is a commonly used coordinate representation method, which follows the right-hand rule to define the direction relationship between the three coordinate axes. The right-handed system consists of three mutually perpendicular coordinate axes, usually marked as the X-axis, the Y-axis, and the Z-axis. Imagine that you stretch out your right hand, and the thumb, index finger, and middle finger are in the same direction as the three coordinate axes. When you point your thumb in the positive direction of the X-axis and your index finger in the positive direction of the Y-axis, according to the right-hand rule, the direction pointed by the middle finger is the positive direction of the Z-axis. In other words, in the right-handed system, if you look from the positive direction of the X-axis to the positive direction of the Y-axis, the Z-axis is upward (zenith direction). Define the train's forward direction as the positive direction of the X-axis and the zenith direction as the positive direction of the Z-axis, so that a right-handed system is formed with the Y-axis, ensuring the direction consistency of the processing process, which is conducive to the subsequent point cloud data processing and analysis.

[0051] Specifically, the positive direction of the X-axis is parallel to the direction in which the train is traveling, which means that when standing at the starting point of the train's route and looking in the direction in which the train is traveling, the X-axis is along the direction of the line of sight.

[0052] Specifically, the Z-axis direction is parallel to the vertical direction, that is, the vertical direction pointing to the sky or the ground. In most cases, the Z-axis usually represents height or depth, so when processing point cloud data, the Z-axis direction can provide information about the height of each point in the point cloud.

[0053] Specifically, in another scheme, the Y-axis direction is perpendicular to the X-axis and the Z-axis respectively. Following the right-hand rule, when your right thumb points to the positive direction of the X-axis and your index finger points to the positive direction of the Z-axis, the direction pointed by your middle finger is the positive direction of the Y-axis. This direction is usually horizontal and related to the width of the train, that is, if you stand on one side of the train, the Y-axis points to the other side of the train.

[0054] Specifically, the X axis is used to locate the length direction of the car, which can help identify the boundaries of the car and the trajectory of the vehicle. The Y axis is used to analyze the width of the car, that is, the lateral characteristics, which is very useful for detecting whether there are abnormal point clouds (such as foreign objects) on both sides. The Z axis is used to determine the height information of each point in the point cloud, which is crucial for identifying the point cloud at the bottom of the car (small Z value) and foreign objects at the bottom of the car (Z value higher than normal).

[0055] Of course, if the train is moving from left to right, the right hand Figure 3 As shown, if the train is moving from right to left, the right hand is Figure 4As shown, the X-axis of the right-hand system is the direction of the train, and the specific direction of the Z-axis is related to the direction of the train, which can be upward, downward, or any other feasible direction. Of course, the X-axis can also be defined as the direction of the train, the Y-axis as the direction of the sky, and the Z-axis as the direction of the carriage perpendicular to the ground.

[0056] Step S202, obtaining point cloud data obtained by scanning the carriage with a laser scanner, wherein the point cloud data is coordinate data of the right-hand coordinate system;

[0057] Specifically, when an empty coal train passes by, a line laser scanner installed above the gantry will scan the entire train to form a scanned point cloud. The point cloud data obtained is as follows: Figure 5 and Figure 6 shown.

[0058] Specifically, when scanning the car, the laser scanner will obtain the three-dimensional coordinate data of each point in the coordinate system, that is, the (X, Y, Z) coordinate value of each point. The laser scanner can measure the distance from each point on the surface of the car to the scanner by emitting a laser beam and receiving the laser reflection. Combined with the position and posture information of the scanner, this distance information can be converted into the precise three-dimensional coordinates of the point in the right-hand coordinate system to form point cloud data.

[0059] Step S203, determining whether there is target point cloud data in the point cloud data whose Z-axis coordinate is greater than or equal to a preset height threshold;

[0060] Specifically, in the right-hand coordinate system, the Z axis represents the vertical direction, so the Z axis coordinate value reflects the height of each point in the point cloud relative to the ground. The preset height threshold is a value set based on experience or specific conditions, which is used to distinguish between normal structures at the bottom of the car and potential foreign objects. The judgment process is: traverse the point cloud data and check whether the Z axis coordinate value of each point is greater than or equal to the preset height threshold. If such points exist, they will be regarded as target point cloud data, which may indicate the presence of foreign objects at the bottom of the car.

[0061] Specifically, the preset height threshold is set based on the average height of the bottom structure of the carriage and the minimum height difference of the foreign objects that may be mounted. The selection of the threshold needs to take into account various situations in actual applications, such as the design of different carriages, the possible debris on the road surface, and the accuracy of laser scanning. Usually, this threshold is fine-tuned through experiments and data analysis to achieve the best foreign object detection effect. Assuming that the average height of the carriage floor under normal conditions is 0.5 meters, considering the possible height of debris on the track, the measurement error of the equipment, and the minimum height of foreign objects, the preset height threshold can be set between 0.6 meters and 0.7 meters. The specific value needs to be carefully adjusted according to the above factors. For example, if the equipment accuracy is high and the minimum height of foreign objects is 10 centimeters, the threshold can be closer to 0.6 meters; if the equipment accuracy is low or the minimum height of foreign objects is 20 centimeters, the threshold may need to be set to 0.7 meters or higher.

[0062] In actual operation, the traversal of point cloud data may involve tens of thousands of points, so it is necessary to perform efficient Z-axis coordinate value checking. A spatial index structure (such as a kd tree or an octree) can be used to speed up this process and reduce calculation time.

[0063] Step S204: when it is determined that the target point cloud data exists in the point cloud data, it is determined that there is a foreign object at the bottom of the carriage.

[0064] Specifically, once points whose Z-axis coordinate values ​​are greater than or equal to a preset height threshold are found in the point cloud data (ie, target point cloud data), a preliminary judgment can be made that foreign objects may exist at the locations of these points.

[0065] Through this embodiment, manual inspection is no longer required. By installing a laser scanner and scanning the car, point cloud data can be obtained. The places with relatively high heights in the point cloud data are where foreign objects are protruding. For example, if there is an extra water cup on the table, the water cup is a foreign object. This solution does not require complex machine learning and deep learning, nor does it require straight line fitting and plane fitting. It greatly simplifies the process of foreign object detection, improves the speed of foreign object detection, and thus improves the efficiency of foreign object detection.

[0066] Specifically, this solution can combine IMU and navigation images for rapid positioning of lunar rovers.

[0067] In the specific implementation process, determining whether there is target point cloud data in the point cloud data whose Z-axis coordinate is greater than or equal to the preset height threshold can be achieved through the following steps: grayscale processing is performed on the point cloud data to obtain a horizontal cross-sectional image; the point cloud data of the frame of the carriage in the horizontal cross-sectional image is removed to obtain a point cloud set; determining whether there is target point cloud data in the point cloud set whose Z-axis coordinate is greater than or equal to the preset height threshold to obtain a determination result.

[0068] In this solution, grayscale processing and border removal can be performed, and height information can be used to identify potential foreign objects, avoiding complex point cloud geometry analysis, thereby further improving the efficiency and simplicity of the solution.

[0069] Specifically, a specific Z-axis height is selected from the original point cloud data to form a horizontal section. This section is actually a two-dimensional point cloud data set, in which each point has X and Y coordinates. Then, the distribution of these points in the section is converted into a grayscale image through some algorithms. In the image, areas with dense point clouds are usually assigned lower grayscale values ​​(i.e. darker colors), while areas with sparse or no point clouds are assigned higher grayscale values ​​(i.e. brighter colors). This can convert the information of the three-dimensional space into a two-dimensional image, which is convenient for subsequent image processing.

[0070] Specifically, once the horizontal cross-sectional image is generated, the next step is to identify and remove the point cloud data representing the border of the car. This is usually done through image processing techniques such as edge detection, identifying those areas with large changes in grayscale values, which may correspond to the edges of the car. Then, the point cloud data of these edge areas is deleted from the horizontal cross-sectional image, leaving a set of point cloud data containing only the interior of the car. The purpose of this is to focus on the point cloud inside the car, avoid the interference of the border on foreign object detection, and improve the accuracy of foreign object recognition.

[0071] Specifically, the last step is to check whether there are points in the remaining point cloud set that are higher than the preset height threshold, that is, the target point cloud data. The preset height threshold is a threshold set based on the height of the normal structure at the bottom of the car and the minimum height difference of possible foreign objects. The algorithm will traverse each point in the point cloud set and check whether its Z coordinate value is greater than or equal to the preset threshold. If such points exist, they are likely to represent foreign objects at the bottom of the car, because under normal circumstances, the points at the bottom of the car should be in a lower height range. Through this judgment, a determination result of whether a foreign object is detected can be obtained.

[0072] Specifically, when converting point cloud data into a grayscale image, it is necessary to determine the grayscale value of each point cloud point in the image. This is usually based on the density, intensity or color information of the point cloud points (if the point cloud is colored). For example, areas with high point cloud density can be assigned lower grayscale values ​​to distinguish them from higher grayscale values ​​in areas with sparse or no point clouds. Specifically, the resolution of the generated grayscale image and the mapping method of the point cloud to the image will affect the efficiency and accuracy of subsequent processing. A high image resolution can capture more detailed point cloud details, but the amount of calculation will also increase. The mapping method of the point cloud to the image (such as uniform sampling, density-based sampling, etc.) will affect the presentation of point cloud information in the image, thereby affecting the accuracy of foreign body detection.

[0073] Specifically, a variety of edge detection algorithms (such as Sobel operator, Canny edge detection algorithm, etc.) can be used to identify the car frame in the horizontal cross-sectional image. These algorithms determine the edge in the image by detecting the gradient change of the grayscale value, thereby helping to determine the outline of the car. In addition to edge detection, image processing techniques such as image segmentation and morphological operations (such as corrosion and expansion) can also be used to further optimize the removal of the border point cloud data to ensure that the removal process is not only accurate, but also can effectively reduce the amount of data for subsequent processing.

[0074] In a specific implementation process, grayscale processing is performed on the point cloud data to obtain a horizontal cross-sectional image, which can be achieved by the following steps: intercepting a point cloud whose length is less than or equal to a preset length from the point cloud data to generate an initial horizontal cross-sectional image; extracting the maximum value of the X axis and the minimum value of the X axis in the initial horizontal cross-sectional image, and extracting the maximum value of the Y axis and the minimum value of the Y axis in the initial horizontal cross-sectional image; creating a blank image, wherein the resolution of the blank image is A×B, and the pixel value in the horizontal direction is A=INT((x max -x min ) / 0.01), the vertical pixel value is B=INT((y max -y min ) / 0.01), x max represents the maximum value of the X axis in the above initial horizontal cross-sectional image, x min represents the minimum value of the X axis in the above initial horizontal cross-section image, y max represents the maximum value of the Y axis in the above initial horizontal cross-sectional image, y min represents the minimum value of the Y axis in the above-mentioned initial horizontal cross-sectional image; the point cloud coordinates in the above-mentioned initial horizontal cross-sectional image are transferred to the above-mentioned blank image with (x min ,y min) is the target coordinate system with the origin as the origin; the point cloud coordinates in the target coordinate system are converted into image coordinates; the grayscale values ​​of the pixels in the image coordinates where the point cloud data exists are adjusted to the first pixel value, and the grayscale values ​​of the pixels in the image coordinates where the point cloud data does not exist are adjusted to the second pixel value, so as to obtain the horizontal cross-sectional image.

[0075] In this solution, a two-dimensional horizontal cross-sectional image is generated from the original three-dimensional point cloud data. The image represents the distribution of the point cloud data at a certain height through different grayscale values. The existence of the point cloud data is indicated by adjusting the grayscale value, which makes subsequent foreign body detection more convenient.

[0076] Specifically, a point cloud with a length of 19 meters can be intercepted from the overall point cloud (the length of the carriage is generally no more than 19 meters), and the highest point can be extracted from this segment of the point cloud. In order to ensure that the special-shaped carriage can also be completely cut, the sampling position is 0.5 meters below the highest point, and the initial horizontal section of the entire carriage is extracted, such as Figure 7 shown.

[0077] Specifically, find the maximum and minimum values ​​of the x- and y-coordinates in the initial horizontal section, that is, x max 、x min ,y max ,y min . Create a blank image with a resolution of A×B, and the horizontal pixel value is A=INT((x max -x min ) / 0.01), the vertical pixel value is B=INT((y max -y min ) / 0.01). Perform coordinate translation and convert the coordinates of all points in the horizontal section to (x min ,y min ) is the origin of the new coordinate system.

[0078] Specifically, the point cloud coordinates (x pi ,y pi ) is converted to image coordinates (x i ,y i ).

[0079]

[0080] The grayscale value of pixels with point clouds is assigned to 0, and the grayscale value of pixels without point clouds is assigned to 255, to form the final horizontal cross-sectional image. Of course, the first pixel value can be 0 or 255, and the second pixel value can also be 0 or 255, and of course, it can also be any other feasible pixel value.

[0081] Specifically, a specific length range (usually not exceeding the length of a carriage) is selected from the overall point cloud, and a horizontal section of the point cloud data within this range is generated. A horizontal section refers to selecting a specific Z-axis height in three-dimensional space, thereby "slicing" the three-dimensional point cloud data into a two-dimensional point cloud data set. The generated horizontal section contains all point cloud data points at the selected height, providing point cloud data with a two-dimensional perspective for subsequent image processing and analysis. The preset length is usually determined based on the size of the carriage. For example, if the average length of the carriage is 18 meters, the preset length may be set to 19 meters to ensure that the point cloud data of the carriage can be fully captured even if the size of the carriage varies slightly.

[0082] Specifically, in the generated horizontal cross-section image, the X-axis and Y-axis represent two directions in the horizontal plane. The maximum and minimum values ​​of the X-axis and the maximum and minimum values ​​of the Y-axis are extracted to determine the range of the point cloud data on the horizontal cross-section, which will help us set the appropriate size when creating a blank image to ensure that all point cloud data points in the initial horizontal cross-section image can be included.

[0083] Specifically, a blank image is a two-dimensional image created based on the range of point cloud data, and its resolution is determined by two parameters, A and B, which represent the width and height of the image, respectively. The horizontal pixel value and the vertical pixel value represent the actual physical size of the width and height of the image, which is usually related to the range of the X-axis and Y-axis of the point cloud data. The size of the created blank image needs to be able to accommodate the entire horizontal cross-section point cloud data, and the choice of resolution will also affect the accuracy and computational efficiency of subsequent image processing.

[0084] Specifically, the point cloud coordinates in the initial horizontal cross-section image are transferred to the created blank image. This transfer process requires redefining the coordinates of the point cloud data to (x min ,y min ) is the origin of the new coordinate system. This is done to ensure alignment between the point cloud data and the blank image so that the point cloud data points can be correctly projected and displayed on the image.

[0085] Specifically, the conversion of point cloud coordinates to image coordinates is the process of mapping points in three-dimensional space (considered as two-dimensional at a specific height) to a two-dimensional image. The position of the point cloud data on the image is marked with different grayscale values. Pixels with point cloud data are assigned a first pixel value (such as dark or 0), indicating that there are actual point cloud data points at that location; pixels without point cloud data are assigned a second pixel value (such as light or 255), indicating blank or no point cloud data. The horizontal cross-sectional image processed in this way can clearly show the distribution of point cloud data, which is convenient for subsequent image processing and analysis, such as compartment segmentation and foreign object detection.

[0086] Specifically, for coordinate transfer, first determine an origin, which is usually the coordinates of the leftmost and bottommost points, that is, (x min ,y min ). This origin will be used as a reference point for the conversion from point cloud data to image data, ensuring that the distribution of points on the image matches the actual point cloud data. Translate all point coordinates in the point cloud data so that the lower left corner of the point cloud is aligned with the origin of the image coordinate system. Since the physical coordinate range of the point cloud data may be much larger than the pixel range of the image, the translated coordinates need to be scaled to adapt to the resolution of the image. Assign a grayscale value to the point cloud data based on its properties (such as point density, reflectivity, etc.). In the simplest case, if there is at least one point cloud data point at a certain pixel position on the image, the grayscale value of the position may be set to a lower value (such as 0, representing black), indicating that there is data; if there is no point cloud data point at this position, its grayscale value is set to a higher value (such as 255, representing white), indicating that it is empty or no data.

[0087] Specifically, when performing coordinate transformation and scaling, the original accuracy of the point cloud data and the resolution of the image need to be taken into account. If the accuracy of the original point cloud data is high, but the image resolution is relatively low, it may result in loss of point cloud details. Conversely, if the image resolution is high, but the accuracy of the point cloud data is insufficient, it may increase unnecessary computational burden.

[0088] Specifically, when performing coordinate transformation, it is necessary to ensure that all point cloud data points fall within the range of the image. If the coordinates of the point cloud data points exceed the size of the image after transformation, pixels should not be allocated to these points on the image to avoid overflow or inaccuracy of the image data.

[0089] Specifically, the grayscale value can be assigned based on the density of the point cloud. For example, if there is more than one point cloud data point at a certain pixel position, the grayscale value of the pixel can be calculated based on the number or density of the point cloud points to reflect the local density of the point cloud.

[0090] Specifically, since the physical coordinates of point cloud data are usually continuous, while the coordinates represented on the image are discrete, decimal coordinate values ​​may be generated during the coordinate conversion process. In order to ensure the accuracy and consistency of the image data, these decimal coordinate values ​​are usually rounded or rounded down, but this may also cause overlap of point cloud data points.

[0091] In the specific implementation process, the point cloud data of the border of the car in the horizontal cross-sectional image is removed to obtain a point cloud set, which can be achieved through the following steps: the horizontal cross-sectional image is segmented to obtain multiple segmented sets; the point cloud data of the border of the car in each of the segmented sets is removed to obtain the point cloud set.

[0092] In this scheme, the horizontal cross-sectional image containing the entire train is segmented into multiple parts, i.e., multiple cut sets, each set roughly corresponding to a carriage. Once the horizontal cross-sectional image is segmented into multiple cut sets containing a single carriage, the next step is to remove the point cloud data of the carriage border from each set. The point cloud data of the border usually represents the side walls and top of the carriage, which may not be important for foreign object detection or contain a large amount of redundant data. Therefore, these unimportant or redundant data can be removed to further improve the accuracy of subsequent foreign object detection.

[0093] Specifically, the segmentation can be completed based on an image processing algorithm. For example, the grayscale value distribution of the image can be used to identify the boundary between carriages, because there are usually obvious gaps or grayscale value mutations between carriages. By identifying these boundary lines, the image can be divided into multiple sub-images, each of which contains the point cloud data of a carriage.

[0094] Specifically, removing the border point cloud data can be done through image processing technology, such as edge detection algorithms, which can identify and highlight edge areas. After identifying the border area, the point cloud data of these areas can be removed from the image, leaving only a point cloud data set containing the bottom of the car and possible foreign objects. In this way, the formed point cloud set is more focused on the features of the bottom of the car, which is convenient for subsequent foreign object detection.

[0095] Specifically, by analyzing the grayscale values ​​of each column in the horizontal cross-sectional image, the gaps between the carriages can be found. If the grayscale value of a column suddenly changes from a low value representing the point cloud data (such as 0, indicating that data exists) to a high value (such as 255, indicating that there is no data), this may be the boundary between the carriages. By continuously checking the grayscale values ​​of multiple columns and identifying stable high grayscale value areas, the boundary position of the carriages can be determined and cut.

[0096] Specifically, in addition to basic grayscale value analysis, more complex image segmentation algorithms can be used to more accurately identify the segmentation position of the carriage, such as segmentation based on region growing, segmentation based on deep learning, etc. These algorithms can identify structural features in the image, such as straight lines, curves, etc., to more accurately determine the outline and boundaries of the carriage.

[0097] Specifically, an edge detection algorithm is used to identify the border of the car, such as Canny edge detection. Once the edges are identified, a mask or bounding box can be created to mark the border area. Then, the point cloud data within these marked areas is removed from the clipping set, leaving a data set that only contains the bottom of the car and potential foreign objects.

[0098] Specifically, in addition to edge detection, the features of point cloud data (such as normal vectors, point cloud density, etc.) can also be used directly to identify the border of the car. For example, points on the side walls of the car usually have normal vectors pointing to the inside of the car, while points on the bottom of the car have normal vectors pointing upwards. By comparing the normal vectors of the points, the point cloud data representing the bottom of the car can be filtered out, thereby removing the border point cloud.

[0099] Specifically, after removing the border point cloud data, the remaining point cloud data can be subjected to density analysis to identify abnormally dense or sparse areas that may represent potential foreign objects or damage. Features (such as shape, size, position, etc.) are extracted from the point cloud and used to further classify possible foreign objects. For example, a clustering algorithm can be used to cluster point cloud data with similar features to form clusters of potential foreign objects. Although the method mentioned in the present invention mainly focuses on the X and Y coordinates of the point cloud, the Z coordinate (height information) is also an important feature of the point cloud data. After removing the border points, the Z coordinate can be further used to identify high points on the bottom of the car, which may be signs of foreign objects.

[0100] In some embodiments, the horizontal cross-sectional image is cut to obtain a plurality of cut sets, which can be specifically achieved by the following steps: obtaining pixel values ​​of the point cloud data in the horizontal cross-sectional image; obtaining a cutting position, wherein the cutting position is a position where the pixel values ​​of all the point cloud data in N consecutive columns are target pixel values, N≥1; and cutting the horizontal cross-sectional image according to the cutting position to obtain a plurality of the cut sets.

[0101] In this scheme, the car boundary line can be automatically identified from the horizontal cross-sectional image, and the image can be segmented into multiple sets containing a single car point cloud data and the car border data can be removed. In this way, different cut sets can be processed subsequently, which speeds up the data processing and facilitates the subsequent foreign object detection.

[0102] Specifically, starting from the leftmost side, the image is judged column by column from left to right. If there are pixels with grayscale values ​​of 0 in the current column, the column is discarded and the next column is judged until there are no pixels with grayscale values ​​of 0 in a certain column, which is marked as S1. Then the next column is judged until a column with no grayscale value of 0 is encountered, and the division of the set S1 is considered complete.

[0103] Specifically, the next column is judged. If there are pixels with grayscale values ​​of 0 in the current column, the column is discarded and the next column is judged. When there are no pixels with grayscale values ​​of 0 in a certain column, it is marked as S2. The next column is judged again until a column with grayscale values ​​of 0 is encountered. The division of the set S2 is considered complete.

[0104] Specifically, the judgment is repeated until all columns are judged. Combined with formula 1, all the cutting position sets {S1, S2, S3, ..., Sn} can be obtained. According to the cutting position set, combined with the position correspondence between pixels and point clouds, all point clouds are cut to complete the extraction of the single carriage point cloud Pi and record the number of carriages.

[0105] Specifically, the grayscale value of each pixel is obtained from the generated horizontal cross-sectional image. Since the point cloud data has been mapped to the image, the pixels with point cloud data will have a lower grayscale value (such as 0), while the pixels without point cloud data will have a higher grayscale value (such as 255). These grayscale values ​​or pixel values ​​will be used for subsequent image processing and analysis.

[0106] Specifically, in the horizontal cross-sectional image, the boundary between cars is usually manifested as several consecutive columns of pixels (continuous pixel columns in the width direction) with high grayscale values, that is, no point cloud data exists, which indicates that there is an "air" gap without point cloud data between these columns. The target pixel value (such as 255) represents the gap between cars or the area without point cloud data. By analyzing the grayscale value of each column of pixels, it is possible to find the columns starting from N consecutive columns (N or more pixel columns) with grayscale values ​​of the target pixel value, which indicates that the point cloud data before and after the column belong to different cars.

[0107] Specifically, once the cutting position, i.e., the dividing line between the carriages, is determined, we can segment the entire horizontal cross-sectional image into multiple sub-images, each of which roughly corresponds to the point cloud data of a single carriage. The cutting process is usually carried out along the vertical direction (the height direction of the image), starting from the cutting position, and the image is segmented into independent sections, each of which is a cutting set. Each cutting set contains the point cloud data of a single carriage. After removing the point cloud data of the carriage border, we can focus more on the features of the bottom of the carriage and possible foreign objects.

[0108] In some embodiments, the point cloud data of the border of the carriage in each of the above-mentioned cut-off sets are removed to obtain the above-mentioned point cloud set, which can be specifically achieved by the following steps: extracting the border point cloud data that meets the preset conditions in the above-mentioned cut-off sets, wherein the above-mentioned preset conditions include that the above-mentioned point cloud data is the maximum value of the X-axis, the above-mentioned point cloud data is the minimum value of the X-axis, the deviation of the normal vector of the above-mentioned point cloud data from the X-axis is less than or equal to the preset angle, the difference between the above-mentioned point cloud data and the maximum value of the X-axis is less than or equal to the preset difference, the difference between the above-mentioned point cloud data and the minimum value of the X-axis is less than or equal to the preset difference, the above-mentioned point cloud data is the maximum value of the Y-axis, the above-mentioned point cloud data is the minimum value of the Y-axis, the deviation of the normal vector of the above-mentioned point cloud data from the Y-axis is less than or equal to the preset angle, the difference between the above-mentioned point cloud data and the maximum value of the Y-axis is less than or equal to the preset difference, and the difference between the above-mentioned point cloud data and the minimum value of the Y-axis is less than or equal to one or more of the above-mentioned preset difference; deleting the above-mentioned border point cloud data that meet the above-mentioned preset conditions in all the above-mentioned cut-off sets to obtain the above-mentioned point cloud set.

[0109] In this scheme, from each cut set (i.e., the point cloud data of each carriage), find and identify the border point cloud data that meets the preset conditions. Once the border point cloud data is determined according to the preset conditions, these data are removed from the cut set to obtain a point cloud set without border interference. The purpose of this is to focus on the point cloud data of the carriage floor, so as to more effectively detect potential foreign objects and reduce the computational burden of subsequent processing.

[0110] Specifically, select the maximum and minimum values ​​of X in Pi respectively, and cluster the point clouds on both sides of the car based on the normal vector and the X value of the point. When the angle deviation between the normal vector and the normal vector of the maximum and minimum values ​​of X is less than 22.5 degrees and the difference in X value is less than 15 cm, the point is classified as the point cloud set on both sides of the car. Traverse all point clouds of Pi until all points are judged. At this point, the point cloud set PC on both sides of the car is determined. Delete the point cloud in Pi that is in the PC set to obtain the remaining point cloud Pi2.

[0111] Specifically, select the maximum and minimum values ​​of Y in Pi2 respectively, and cluster the point clouds on both sides of the car based on the normal vector and the Y value of the point. When the angle deviation between the normal vector and the maximum and minimum values ​​of Y is less than 22.5 degrees and the difference in Y value is less than 15 cm, the point is classified as the point cloud set at both ends of the car. Traverse all point clouds of Pi2 until all points are judged. At this point, the point cloud set PC2 on both sides of the car is determined. Delete the point cloud in Pi2 that is in the PC2 set to obtain the final point cloud PF that needs to be sliced.

[0112] Specifically, starting from the lower left corner point in PF, the point cloud of a single carriage is cut along the planes parallel to the XZ plane and the YZ plane at a spacing of 1 cm to form multiple cross-sectional (parallel to the XZ plane) and longitudinal (parallel to the YZ plane) point cloud slices.

[0113] Specifically, first start the point cloud processing of the cross section. Determine the foreign body points of each section. First, take the first cross section H1, count the Z values ​​of each point, and record the minimum Z value Zmin. Start judging point by point from the first point of the section. If the difference between the Z value and Zmin is greater than 2 cm, the point is considered to be a foreign body point and is included in the foreign body point set W1. Take out the next cross section Hi, count the Z values ​​of each point, record the minimum Z value Zmini, and perform the foreign body point judgment operation. After all cross-section judgments are completed, repeat the foreign body judgment steps to judge the foreign body points of the longitudinal section to obtain the final foreign body point set W2. Merge the foreign body point sets W1 and W2 to obtain the set W of all foreign body points, and obtain the final foreign body extraction result of this car. Repeat the judgment of each car to complete the foreign body extraction of each car.

[0114] Specifically, the point cloud data is the maximum or minimum value of the X-axis: this represents that the point is located at the edges of both sides of the car body. The deviation of the normal vector of the point cloud data from the X-axis is less than or equal to the preset angle, ensuring that the normal vectors of these points are roughly perpendicular to the side walls of the car body, which helps to exclude points at the bottom of the car body or irrelevant points. The difference between the point cloud data and the maximum or minimum value of the X-axis is less than or equal to the preset difference, which further limits the range of the border point cloud data and identifies those points that are closely attached to the side walls of the car body as border data. The selection of the preset difference should take into account the average structural thickness of the car body and the accuracy of the point cloud data. Similarly, for the direction of the Y-axis, the above conditions also apply to identifying the front and back of the car body.

[0115] Specifically, the setting of preset conditions needs to be adjusted according to the accuracy of point cloud data, the structural characteristics of the car body and the specific needs of foreign object detection. For example, the preset angle should take into account that the normal vector of the car body edge may be affected by point cloud data noise or structural details, while the preset difference should be set according to the average thickness of the car body and the possible size of foreign objects.

[0116] Preprocessing of point cloud data,Before identifying the bounding box point cloud data, it may be necessary to preprocess the point cloud data, such as filtering or smoothing, to reduce the impact of noise and outliers and improve the accuracy of matching the preset conditions.

[0117] Optimization of normal vector calculation,Calculating the normal vector of each point is a computationally intensive task, especially on large point cloud datasets.,The efficiency of this step can be improved by optimizing the algorithm or utilizing,efficient computing frameworks (such as GPU acceleration) to ensure the real-time,performance of the entire processing flow.

[0118] Removal algorithm of border point cloud data,Removal of border point cloud data can be achieved through a variety of algorithms, such as methods based on region growing or connected component analysis.,After identifying all border points that meet the preset conditions,,these points can be quickly deleted by creating a mask or an index list of,point cloud data points, avoiding the need to traverse the entire point cloud data set,to check each point one by one.

[0119] Visualization and verification of point cloud data,After deleting the border point cloud data, the processed point cloud data can be visualized, and the accuracy and completeness of the processing results can be checked by 3D rendering or 2D projection. This step is particularly important for debugging algorithm parameters and verifying processing results.

[0120] The prerequisite for foreign object detection is to remove the border point cloud data, so that the point cloud set can be more focused on the characteristics of the car floor, which provides a more accurate and clear data basis for subsequent foreign object detection. Without the interference of border points, the foreign object detection algorithm can more accurately identify abnormal point cloud clusters on the car floor, improving the efficiency and accuracy of detection.

[0121] In some embodiments, after determining that there is a foreign object at the bottom of the above-mentioned carriage, the above-mentioned method further includes the following steps: generating an alarm message when it is determined that there is a foreign object at the bottom of the above-mentioned carriage; and sending the above-mentioned alarm message to a target terminal.

[0122] In this solution, if a point cloud cluster that does not conform to normal characteristics is identified at the bottom of the carriage, that is, it is judged that there is a foreign object, an alarm message will be automatically generated. The alarm message will be sent to the target terminal in a timely manner so that the staff can promptly understand that there is a foreign object at the bottom of the carriage.

[0123] Specifically, the alarm information usually contains key information such as the carriage number where the foreign object was found, the possible location, type or size of the foreign object, and the exact time of detection. The generated alarm information needs to be sent to the personnel or system responsible for monitoring and response, that is, the target terminal, in a timely manner so that they can take prompt action. The target terminal can be a computer in the monitoring center, a mobile device, a dedicated alarm system, or a device that communicates directly with the train operator. Through real-time communication technologies such as wireless networks and satellite communications, alarm information can be quickly and reliably transmitted to the target terminal.

[0124] Specifically, the alarm message should contain as detailed and useful information as possible, such as the specific location of the foreign object (coordinates in the car), the description of the foreign object (shape, size, potential type), the detection time, the car number or location information, and any additional data that will help with response and processing.

[0125] Specifically, ensuring that alarm information can be sent to the target terminal safely and reliably requires the use of appropriate communication protocols and technologies. For example, encryption technology can be used to protect the security of information during transmission, and redundant communication paths or multiple target terminals can be used to improve the reliability of information transmission and prevent alarm delays or losses caused by signal interruptions or equipment failures.

[0126] Specifically, after receiving the alarm information, the target terminal should immediately initiate the response mechanism, which may include notifying maintenance personnel, automatically adjusting the train operation plan, triggering remote or automatic detailed inspection procedures, etc. Real-time feedback and processing based on alarm information is an important part of improving the overall efficiency and safety of the foreign object detection system.

[0127] Specifically, in the future, deep learning or other artificial intelligence technologies can be combined to further analyze the type and potential impact of foreign objects, and provide more specific guidance to the response team, such as whether it is necessary to stop and inspect immediately, whether the foreign objects can be removed remotely by robotic arms, etc., thereby improving the level of intelligence in handling foreign objects.

[0128] Specifically, the generation and sending of alarm information can be integrated into a wider train early warning system, working in conjunction with other safety systems such as track status detection and weather condition monitoring to provide comprehensive safety protection for train operations.

[0129] Specifically, the alarm information should not only be sent to the target terminal, but also synchronized to the train's command center so that decisions and scheduling can be made from a global perspective of train operation. The command center can adjust the train's operation plan to ensure safety based on the received alarm information and other relevant data.

[0130] In summary, the solution of the present application selects a horizontal cross section of the train at a specific position, and converts the cross-sectional point cloud into a two-dimensional image through coordinate transformation. Then, the car segmentation position is determined by each column of grayscale values, and the point cloud of the entire train car is divided into single-carriage point clouds according to the segmentation position. After that, the redundant data in the single-carriage point cloud is removed by point cloud clustering, the point cloud at the bottom of the car is retained, and a point cloud profile is formed. Finally, the extraction of foreign objects in the car is completed by the difference between the Z value of each point in the cross-sectional point cloud and the Z value of the foreign object-free area. The invention does not require machine learning and training processes, does not require straight line and plane fitting operations, does not rely on GPU computing resources, and also avoids the situation where the program takes too long or foreign objects are missed.

[0131] In summary, the solution of the present application solves the problem of point cloud segmentation of a single carriage and foreign object extraction inside the carriage during the process of foreign object extraction from the point cloud of the entire carriage. Compared with the existing technical solutions, the technical solution of the present invention has the following obvious advantages:

[0132] 1. This method can achieve good foreign body extraction results without machine learning and deep learning, and does not require GPU resources;

[0133] 2. Compared with the point cloud segmentation method, the solution in this paper does not require straight line or plane fitting, which to some extent avoids the situation where the program takes too long or foreign objects are missed;

[0134] 3. This method simplifies part of the point cloud clustering process into an image processing process, discards the Z value of the point cloud, and speeds up the processing speed of the program;

[0135] 4. When extracting foreign objects, the carriage number is recorded for subsequent processing. Compared with traditional methods, no additional sensors are required for carriage identification and other operations, which reduces the cost of the foreign object detection system.

[0136] The embodiment of the present application also provides a foreign object determination device at the bottom of a carriage. It should be noted that the foreign object determination device at the bottom of a carriage in the embodiment of the present application can be used to execute the foreign object determination method for the bottom of a carriage provided in the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0137] The following is an introduction to the foreign object determination device at the bottom of the vehicle compartment provided in an embodiment of the present application.

[0138] Figure 8 is a structural block diagram of a foreign object determination device at the bottom of a vehicle compartment according to an embodiment of the present application. Figure 8 As shown, the device comprises:

[0139] An establishing unit 10 is used to establish a right-handed coordinate system of the carriage, wherein the positive direction of the X-axis of the right-handed coordinate system is parallel to the forward direction of the train, the Z-axis of the right-handed coordinate system is parallel to the vertical direction, and the Y-axis of the right-handed coordinate system is perpendicular to the X-axis and the Z-axis respectively;

[0140] An acquisition unit 20 is used to acquire point cloud data obtained by scanning the carriage with a laser scanner, wherein the point cloud data is coordinate data of the right-hand coordinate system;

[0141] The first determination unit 30 is used to determine whether there is target point cloud data in which the Z-axis coordinate is greater than or equal to a preset height threshold in the above point cloud data;

[0142] The second determining unit 40 is used to determine whether there is a foreign object at the bottom of the carriage when it is determined that the target point cloud data exists in the point cloud data.

[0143] Through this embodiment, manual inspection is no longer required. By installing a laser scanner and scanning the car, point cloud data can be obtained. The places with relatively high heights in the point cloud data are where foreign objects are protruding. For example, if there is an extra water cup on the table, the water cup is a foreign object. This solution does not require complex machine learning and deep learning, nor does it require straight line fitting and plane fitting. It greatly simplifies the process of foreign object detection, improves the speed of foreign object detection, and thus improves the efficiency of foreign object detection.

[0144] During the specific implementation process, the first determination unit includes a first processing module, a second processing module and a determination module. The first processing module is used to perform grayscale processing on the above-mentioned point cloud data to obtain a horizontal cross-sectional image; the second processing module is used to remove the above-mentioned point cloud data of the border of the above-mentioned car in the above-mentioned horizontal cross-sectional image to obtain a point cloud set; the determination module is used to determine whether there is the above-mentioned target point cloud data in the Z-axis coordinate greater than or equal to the above-mentioned preset height threshold in the above-mentioned point cloud set to obtain a determination result.

[0145] In this solution, grayscale processing and border removal can be performed, and height information can be used to identify potential foreign objects, avoiding complex point cloud geometry analysis, thereby further improving the efficiency and simplicity of the solution.

[0146] In the specific implementation process, the first processing module includes a generation submodule, an extraction submodule, a creation submodule, a transfer submodule, a conversion submodule and an adjustment submodule. The generation submodule is used to intercept a point cloud whose length is less than or equal to a preset length from the above point cloud data to generate an initial horizontal section image; the extraction submodule is used to extract the maximum value of the X axis and the minimum value of the X axis in the above initial horizontal section image, and extract the maximum value of the Y axis and the minimum value of the Y axis in the above initial horizontal section image; the creation submodule is used to create a blank image, wherein the resolution of the blank image is A×B, and the pixel value in the horizontal direction is A=INT((x max -x min ) / 0.01), the vertical pixel value is B=INT((y max -y min ) / 0.01), x max represents the maximum value of the X axis in the above initial horizontal cross-sectional image, x min represents the minimum value of the X axis in the above initial horizontal cross-section image, y max represents the maximum value of the Y axis in the above initial horizontal cross-sectional image, y min represents the minimum value of the Y axis in the above-mentioned initial horizontal cross-sectional image; the transfer submodule is used to transfer the point cloud coordinates in the above-mentioned initial horizontal cross-sectional image to the above-mentioned blank image in the form of (x min ,y min) is in the target coordinate system with the origin as the origin; the conversion submodule is used to convert the point cloud coordinates in the target coordinate system into image coordinates; the adjustment submodule is used to adjust the grayscale value of the pixel in the image coordinate where the point cloud data exists to the first pixel value, and adjust the grayscale value of the pixel in the image coordinate where the point cloud data does not exist to the second pixel value, so as to obtain the horizontal cross-sectional image.

[0147] In this solution, a two-dimensional horizontal cross-sectional image is generated from the original three-dimensional point cloud data. The image represents the distribution of the point cloud data at a certain height through different grayscale values. The existence of the point cloud data is indicated by adjusting the grayscale value, which makes subsequent foreign body detection more convenient.

[0148] During the specific implementation process, the second processing module includes a cutting submodule and a removal submodule. The cutting submodule is used to cut the above-mentioned horizontal cross-sectional image to obtain multiple cutting sets; the removal submodule is used to remove the above-mentioned point cloud data of the border of the above-mentioned carriage in each of the above-mentioned cutting sets to obtain the above-mentioned point cloud set.

[0149] In this scheme, the horizontal cross-sectional image containing the entire train is segmented into multiple parts, i.e., multiple cut sets, each set roughly corresponding to a carriage. Once the horizontal cross-sectional image is segmented into multiple cut sets containing a single carriage, the next step is to remove the point cloud data of the carriage border from each set. The point cloud data of the border usually represents the side walls and top of the carriage, which may not be important for foreign object detection or contain a large amount of redundant data. Therefore, these unimportant or redundant data can be removed to further improve the accuracy of subsequent foreign object detection.

[0150] In some embodiments, the cutting submodule is also used to obtain the pixel values ​​of the above-mentioned point cloud data in the above-mentioned horizontal cross-sectional image; the cutting submodule is also used to obtain the cutting position, wherein the above-mentioned cutting position is the position where the pixel values ​​of all the above-mentioned point cloud data in N consecutive columns are the target pixel values, N≥1; the cutting submodule is also used to cut the above-mentioned horizontal cross-sectional image according to the above-mentioned cutting position to obtain the above-mentioned multiple cutting sets.

[0151] In this scheme, the car boundary line can be automatically identified from the horizontal cross-sectional image, and the image can be segmented into multiple sets containing a single car point cloud data and the car border data can be removed. In this way, different cut sets can be processed subsequently, which speeds up the data processing and facilitates the subsequent foreign object detection.

[0152] In some embodiments, the removal submodule is also used to extract the border point cloud data that meets the preset conditions in the above-mentioned cutting set, wherein the above-mentioned preset conditions include that the above-mentioned point cloud data is the maximum value of the X-axis, the above-mentioned point cloud data is the minimum value of the X-axis, the deviation of the normal vector of the above-mentioned point cloud data from the X-axis is less than or equal to the preset angle, the difference between the above-mentioned point cloud data and the maximum value of the X-axis is less than or equal to the preset difference, the difference between the above-mentioned point cloud data and the minimum value of the X-axis is less than or equal to the preset difference, the above-mentioned point cloud data is the maximum value of the Y-axis, the above-mentioned point cloud data is the minimum value of the Y-axis, the deviation of the normal vector of the above-mentioned point cloud data from the Y-axis is less than or equal to the preset angle, the difference between the above-mentioned point cloud data and the maximum value of the Y-axis is less than or equal to the preset difference, and the difference between the above-mentioned point cloud data and the minimum value of the Y-axis is less than or equal to one or more of the preset difference; the removal submodule is also used to delete the above-mentioned border point cloud data that meets the above-mentioned preset conditions in all the above-mentioned cutting sets to obtain the above-mentioned point cloud set.

[0153] In this scheme, from each cut set (i.e., the point cloud data of each carriage), find and identify the border point cloud data that meets the preset conditions. Once the border point cloud data is determined according to the preset conditions, these data are removed from the cut set to obtain a point cloud set without border interference. The purpose of this is to focus on the point cloud data of the carriage floor, so as to more effectively detect potential foreign objects and reduce the computational burden of subsequent processing.

[0154] In some embodiments, the above-mentioned device also includes a generating unit and a sending unit. The generating unit is used to generate an alarm message after determining that there is a foreign object at the bottom of the above-mentioned carriage; the sending unit is used to send the above-mentioned alarm message to the target terminal.

[0155] In this solution, if a point cloud cluster that does not conform to normal characteristics is identified at the bottom of the carriage, that is, it is judged that there is a foreign object, an alarm message will be automatically generated. The alarm message will be sent to the target terminal in a timely manner so that the staff can promptly understand that there is a foreign object at the bottom of the carriage.

[0156] The foreign object determination device at the bottom of the carriage includes a processor and a memory. The establishment unit, the acquisition unit, the first determination unit, the second determination unit, etc. are all stored in the memory as program units, and the processor executes the program units stored in the memory to implement corresponding functions. The modules are all located in the same processor; or, the modules are located in different processors in any combination.

[0157] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be provided, and the problem of low efficiency of foreign body detection at the bottom of train carriages in the prior art can be solved by adjusting kernel parameters.

[0158] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0159] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored program. When the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method for determining foreign objects at the bottom of a vehicle compartment.

[0160] An embodiment of the present invention provides a processor, and the processor is used to run a program, wherein the method for determining foreign objects at the bottom of a vehicle compartment is executed when the program is run.

[0161] The embodiment of the present invention provides a device, which includes a processor, a memory, and a program stored in the memory and executable on the processor, and when the processor executes the program, at least the steps of the method for determining a foreign object at the bottom of a vehicle compartment are implemented. The device in this article can be a server, a PC, a PAD, a mobile phone, etc.

[0162] A computer program product includes a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for determining foreign objects at the bottom of a vehicle compartment in each embodiment of the present application are implemented.

[0163] The present application also provides a vehicle compartment detection system, comprising one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a method for executing any one of the above-mentioned foreign object determination methods at the bottom of the vehicle compartment.

[0164] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0165] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0166] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0167] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0168] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0169] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0170] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0171] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0172] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0173] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0174] 1) The method for determining foreign objects at the bottom of the car compartment of the present application does not require manual inspection. By installing a laser scanner and scanning the car compartment, point cloud data can be obtained. The places with relatively high heights in the point cloud data are where foreign objects protrude. For example, if there is an extra water cup on the table, the water cup is a foreign object. This solution does not require complex machine learning and deep learning, nor does it require straight line fitting and plane fitting. It greatly simplifies the process of foreign object detection, improves the speed of foreign object detection, and thus improves the efficiency of foreign object detection.

[0175] 2) The foreign object determination device at the bottom of the car of the present application does not require manual detection. By installing a laser scanner, point cloud data can be obtained by scanning the car. The places with relatively high heights in the point cloud data are where foreign objects protrude. For example, if there is an extra water cup on the table, the water cup is a foreign object. This solution does not require complex machine learning and deep learning, nor does it require straight line fitting and plane fitting. It greatly simplifies the process of foreign object detection, improves the speed of foreign object detection, and thus improves the efficiency of foreign object detection.

[0176] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for determining foreign matter at the bottom of a vehicle compartment, characterized in that: include: Establish a right-handed coordinate system of the carriage, wherein the positive direction of the X-axis of the right-handed coordinate system is parallel to the forward direction of the train, the Z-axis of the right-handed coordinate system is parallel to the vertical direction, and the Y-axis of the right-handed coordinate system is perpendicular to the X-axis and the Z-axis respectively; Acquire point cloud data obtained by scanning the carriage with a laser scanner, wherein the point cloud data is coordinate data of the right-hand coordinate system; Determine whether there is target point cloud data in the point cloud data whose Z-axis coordinate is greater than or equal to a preset height threshold; When it is determined that the target point cloud data exists in the point cloud data, it is determined that there is a foreign object at the bottom of the carriage.

2. The method according to claim 1, characterized in that: Determining whether there is target point cloud data in the point cloud data whose Z-axis coordinate is greater than or equal to a preset height threshold includes: Performing grayscale processing on the point cloud data to obtain a horizontal cross-sectional image; Removing the point cloud data of the frame of the carriage in the horizontal cross-sectional image to obtain a point cloud set; Determine whether there is target point cloud data in the point cloud set whose Z-axis coordinate is greater than or equal to the preset height threshold, and obtain a determination result.

3. The method according to claim 2, characterized in that Grayscale processing is performed on the point cloud data to obtain a horizontal cross-sectional image, including: Extracting point clouds whose length is less than or equal to a preset length from the point cloud data to generate an initial horizontal cross-sectional image; Extracting the maximum value of the X axis and the minimum value of the X axis in the initial horizontal cross-sectional image, and extracting the maximum value of the Y axis and the minimum value of the Y axis in the initial horizontal cross-sectional image; Create a blank image, wherein the resolution of the blank image is A×B, and the pixel value in the horizontal direction is A=INT((x max -x min ) / 0.01), the vertical pixel value is B=INT((y max -y min ) / 0.01), x max represents the maximum value of the X axis in the initial horizontal cross-sectional image, x min represents the minimum value of the X axis in the initial horizontal cross-sectional image, y max represents the maximum value of the Y axis in the initial horizontal cross-sectional image, y min represents the minimum value of the Y axis in the initial horizontal cross-sectional image; The point cloud coordinates in the initial horizontal cross-section image are transferred to the blank image with (x min ,y min ) is the target coordinate system with the origin; Convert the point cloud coordinates in the target coordinate system into image coordinates; The grayscale value of the pixel where the point cloud data exists in the image coordinates is adjusted to a first pixel value, and the grayscale value of the pixel where the point cloud data does not exist in the image coordinates is adjusted to a second pixel value, so as to obtain the horizontal cross-sectional image.

4. The method according to claim 2, characterized in that: The point cloud data of the frame of the carriage in the horizontal cross-sectional image is removed to obtain a point cloud set, including: Cutting the horizontal cross-sectional image to obtain a plurality of cut sets; The point cloud data of the border of the carriage in each of the cut sets is removed to obtain the point cloud set.

5. The method according to claim 4, characterized in that The horizontal cross-sectional image is cut to obtain a plurality of cut sets, including: Acquire pixel values ​​of the point cloud data in the horizontal cross-sectional image; Acquire a cutting position, wherein the cutting position is a position where the pixel values ​​of all the point cloud data in N consecutive columns are the target pixel values, and N≥1; The horizontal cross-sectional image is segmented according to the cutting position to obtain a plurality of segmentation sets.

6. The method according to claim 4, characterized in that The point cloud data of the border of the carriage in each of the cut sets is removed to obtain the point cloud set, including: Extracting the frame point cloud data that meets the preset conditions in the clipping set, wherein the preset conditions include one or more of the following: the point cloud data is the maximum value of the X-axis, the point cloud data is the minimum value of the X-axis, the deviation of the normal vector of the point cloud data from the X-axis is less than or equal to a preset angle, the difference between the point cloud data and the maximum value of the X-axis is less than or equal to a preset difference, the difference between the point cloud data and the minimum value of the X-axis is less than or equal to the preset difference, the point cloud data is the maximum value of the Y-axis, the point cloud data is the minimum value of the Y-axis, the deviation of the normal vector of the point cloud data from the Y-axis is less than or equal to the preset angle, the difference between the point cloud data and the maximum value of the Y-axis is less than or equal to the preset difference, and the difference between the point cloud data and the minimum value of the Y-axis is less than or equal to the preset difference; The frame point cloud data satisfying the preset condition in all the clipping sets are deleted to obtain the point cloud set.

7. The method according to any one of claims 1 to 6, characterized in that After determining that there is a foreign object at the bottom of the carriage, the method further includes: If it is determined that there is a foreign object at the bottom of the carriage, generating an alarm message; The alarm information is sent to the target terminal.

8. A foreign object determination device at the bottom of a carriage, characterized in that: include: An establishing unit is used to establish a right-handed coordinate system of the carriage, wherein the positive direction of the X-axis of the right-handed coordinate system is parallel to the forward direction of the train, the Z-axis of the right-handed coordinate system is parallel to the vertical direction, and the Y-axis of the right-handed coordinate system is perpendicular to the X-axis and the Z-axis respectively; An acquisition unit, used for acquiring point cloud data obtained by scanning the carriage with a laser scanner, wherein the point cloud data is coordinate data of the right-hand coordinate system; A first determining unit is used to determine whether there is target point cloud data in which the Z-axis coordinate is greater than or equal to a preset height threshold in the point cloud data; The second determining unit is used to determine that there is a foreign object at the bottom of the carriage when it is determined that the target point cloud data exists in the point cloud data.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method for determining a foreign object at the bottom of a vehicle compartment as claimed in any one of claims 1 to 7 are implemented.

10. A carriage detection system, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a method for executing the foreign object determination method at the bottom of a vehicle compartment as described in any one of claims 1 to 7.

Images