Automatic hooking method and automatic unhooking method of railway vehicle and electronic equipment

Abstract

The invention provides an automatic hooking method and an automatic unhooking method of a railway vehicle and electronic equipment. By combining the laser radar and the three-dimensional point cloud data, the automatic hooking process of the railway vehicle is accurately controlled, and the method has remarkable beneficial effects. Firstly, three-dimensional point cloud data are continuously collected through a laser radar, the position of a target car coupler can be monitored in real time, interference of environmental factors such as illumination and rain and snow is avoided, and the defect that a traditional visual system is insufficient in recognition in complex weather is overcome. And secondly, by calculating the distance deviation data between the car couplers and adjusting the running speed according to the preset speed control strategy, it is ensured that the first rail vehicle can accurately approach the target car coupler at the proper speed, and the risk of collision or dislocation is reduced. In addition, after it is judged that the preset hooking distance is reached, hooking closing operation is automatically executed, automation and high efficiency of the hooking process are guaranteed, the requirement for manual intervention is reduced, and operation safety and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic control of rail vehicles, and particularly to an automatic coupling method, an automatic uncoupling method and an electronic device for rail vehicles. Background Art

[0002] In the current fields of rail transit and industrial transportation, the coupling operation between vehicles mostly relies on manual completion or semi-automatic equipment assistance. The traditional coupling and uncoupling operations not only have low efficiency, but also have great potential safety hazards. Especially in high-temperature, high-noise, dusty or toxic environments, manual operations are extremely prone to misoperations and personal injuries.

[0003] In the related art, during the process of automatic coupling of two rail vehicles, there is a technical problem that the coupler positioning is not accurate enough. Summary of the Invention

[0004] The purpose of the present invention is to overcome the above technical deficiencies, and provide an automatic coupling method, an automatic uncoupling method and an electronic device for rail vehicles, so as to solve the technical problem that the coupler positioning is not accurate enough during the process of automatic coupling of two rail vehicles in the related art.

[0005] To achieve the above technical purpose, the present invention adopts the following technical solutions: In a first aspect, the present invention provides an automatic coupling method for a rail vehicle, including: Controlling a first rail vehicle to travel at a preset first speed, and continuously collecting three-dimensional point cloud data in front of the first rail vehicle through a preset lidar during the travel; wherein, in front of the first rail vehicle refers to in front of the body coupler of the first rail vehicle; the lidar is arranged on the first rail vehicle and is on the same side as the body coupler; Based on the three-dimensional point cloud data, determining whether there is a target coupler; wherein the target coupler is configured on one side of a second rail vehicle relative to the body coupler of the first rail vehicle; In the case of determining that there is a target coupler, calculating the docking deviation data between the target coupler and the body coupler; wherein the docking deviation data at least includes the docking distance deviation data; Based on the distance deviation data and a preset speed control strategy, controlling the travel speed of the first rail vehicle so that the first rail vehicle approaches the second rail vehicle at a speed not higher than and not equal to the first speed; In the case of determining that the preset coupling distance is reached, performing coupling closure.

[0006] Further, the step of determining whether there is a target coupler based on the three-dimensional point cloud data includes: Perform feature extraction on the three-dimensional point cloud data to obtain feature data for characterizing the coupler features; Based on the feature data for characterizing the coupler features, determine whether there is a target coupler in front of the main coupler.

[0007] Furthermore, the docking deviation data further includes: docking angle deviation data; The step of calculating the docking deviation data between the target coupler and the main coupler in the case where it is determined that there is a target coupler includes: Extract target three-dimensional point cloud data based on the latest three-dimensional point cloud data; wherein, the target three-dimensional point cloud data is the three-dimensional point cloud data representing the area where the target coupler is located; Perform a preset surface fitting algorithm on the target three-dimensional point cloud data to generate a fitted quadratic surface model; Based on the quadratic surface model, calculate the center point coordinates and normal vector direction of the target coupler; Based on the pre-calibrated center point coordinates and normal vector direction of the main coupler and the center point coordinates and normal vector direction of the target coupler, calculate the distance deviation data and the docking angle deviation data.

[0008] Furthermore, the step of controlling the traveling speed of the first rail vehicle based on the distance deviation data and a preset speed control strategy includes: In the case where it is determined that there is a target coupler, switch the first speed to the second speed and control the first rail vehicle to travel at a constant speed at the second speed; wherein, the second speed is lower than the first speed; Determine whether the distance deviation data reaches a preset speed switching threshold; In the case where the preset speed switching threshold is reached, switch the second speed to the third speed and control the first rail vehicle to travel at the third speed; wherein, the third speed is lower than the second speed.

[0009] Furthermore, based on the distance deviation data and a preset speed control strategy, control the traveling speed of the first rail vehicle so that before the first rail vehicle approaches the second rail vehicle at a speed not higher than and not equal to the first speed, and in the case where the coupler is a Janney coupler, the method further includes: Send a control signal to the power device of the main coupler to open the main coupler so that the main coupler is in a state of waiting for connection; The step of performing hook closing in the case where it is determined that the preset hooking distance is reached includes: In the case where it is determined that the preset hooking distance is reached, control the first rail vehicle and the second rail vehicle to touch at a speed not higher than the third speed; After detecting the occurrence of a touch, three-dimensional point cloud data of the hook area is collected by a preset lidar; For the three-dimensional point cloud data of the hook area, a point cloud voxel block of the locking pin area of the Janney coupler is extracted; Based on the extracted point cloud voxel block of the locking pin area and a preset locking pin landing model, an initial judgment on whether the coupling is successful is performed; In the case where the initial judgment is that the coupling is successful, a preset final judgment strategy is executed to finally judge whether the coupling is successful.

[0010] Further, the step of performing an initial judgment on whether the coupling is successful based on the extracted point cloud voxel block of the locking pin area and a preset locking pin landing model includes: The point cloud voxel block of the locking pin area is matched and compared with the pre-stored reference point cloud spatial distribution in the locking pin landing model to obtain the spatial alignment relationship between the two; Based on the spatial alignment relationship, a point cloud cluster corresponding to the locking pin end point is extracted from the point cloud voxel block, and the center coordinates of the point cloud cluster are calculated; It is judged whether the center coordinates of the point cloud cluster fall within the three-dimensional tolerance range of the hook head pin hole area calibrated in the locking pin landing model, and it is judged whether the absolute value of the vertical direction coordinate deviation is within a preset vertical deviation range; In the case where both are satisfied, it is judged that the locking pin has landed correctly, and the initial judgment is that the coupling is successful.

[0011] Further, the step of executing a preset final judgment strategy to judge whether the coupling is successful in the case where the initial judgment is that the coupling is successful includes: Controlling the first rail vehicle to move reversely a preset distance along the driving direction, and during the reverse movement, continuously collecting three-dimensional point cloud data of the locking pin area through the lidar; Based on the three-dimensional point cloud data collected during the reverse movement, a real-time point cloud cluster corresponding to the locking pin end point is extracted, and its real-time center coordinates are calculated; According to the continuous change amount of the real-time center coordinates, it is judged whether the position of the locking pin meets a preset stability condition; wherein, the preset stability condition includes: the horizontal displacement change amount of the real-time center coordinates of the locking pin end point does not exceed the horizontal deviation range; the vertical displacement change amount does not exceed the vertical deviation range; If the real-time center coordinates meet the preset stability condition, it is finally determined that the coupling is successful.

[0012] In a second aspect, the present invention provides an automatic uncoupling method for a rail vehicle, including: In response to the received uncoupling instruction, a control signal is sent to the power device of the main coupler of the first rail vehicle to drive the power device to perform a coupler unlocking operation; wherein, the main coupler and the target coupler are in a coupled state; wherein, the first rail vehicle has performed a coupling operation between the main coupler and the target coupler by adopting the above-mentioned automatic coupling method for rail vehicles, so that the main coupler and the target coupler are in a coupled state; During the execution of the coupler unlocking operation, three-dimensional point cloud data of the coupling area is collected by a preset lidar, and the real-time point cloud voxel block of the locking pin area is extracted; Based on the real-time point cloud voxel block and a preset locking pin not-in-place model for matching and comparison, the result of the matching and comparison is obtained; wherein, the locking pin not-in-place model includes the reference point cloud distribution characteristics after the locking pin is pulled out; Based on the result of the matching and comparison, it is judged whether the locking pin has been successfully pulled out; In the case where it is determined that the locking pin has been successfully pulled out, the first rail vehicle is controlled to move backward until the lidar detects that the coupling separation distance is greater than a preset safety distance threshold, and then the uncoupling is confirmed to be completed.

[0013] Further, the step of judging whether the locking pin has been successfully pulled out based on the result of the matching and comparison includes: If the point cloud voxel block of the locking pin area does not cover the predefined locking pin feature area in the locking pin not-in-place model, or the center coordinate of the point cloud cluster corresponding to the locking pin end deviates from the center of the coupling hole position by more than a preset uncoupling tolerance range, it is determined that the locking pin has been successfully pulled out; If the center coordinate of the point cloud cluster corresponding to the locking pin end is still within the preset uncoupling tolerance range of the coupling hole position area, it is determined that the uncoupling is not completed, and an unlocking abnormal alarm signal is generated.

[0014] In a third aspect, the present invention provides an electronic device, including: a memory, and one or more processors communicatively connected to the memory; instructions executable by the one or more processors are stored in the memory, and when the instructions are executed by the one or more processors, the one or more processors are enabled to implement the above method.

[0015] Beneficial effects: By combining lidar and 3D point cloud data, the present invention precisely controls the automatic coupling process of rail vehicles, with significant beneficial effects. First, by continuously collecting 3D point cloud data through lidar, the position of the target coupler can be monitored in real time, without being interfered by environmental factors such as light, rain, and snow, overcoming the recognition deficiencies of traditional vision systems in complex weather. Second, by calculating the distance deviation data between the couplers and adjusting the driving speed according to the preset speed control strategy, it ensures that the first rail vehicle can accurately approach the target coupler at an appropriate speed, reducing the risk of collision or misalignment. In addition, after determining that the preset coupling distance is reached, the coupling closing operation is automatically executed, ensuring the automation and efficiency of the coupling process, reducing the need for manual intervention, and enhancing the operation safety and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a schematic flowchart of an automatic coupling method for a rail vehicle provided by an embodiment of the present invention; Figure 2 is a schematic flowchart of an automatic uncoupling method for a rail vehicle provided by an embodiment of the present invention; Figure 3 is a block diagram of an electronic device adopted by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] In order to enable those skilled in the art of the present technology to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present application.

[0018] In the related art, in the field of rail transit, automatic coupling technology is applied to improve transportation efficiency and reduce manual operations. However, existing automatic coupling systems face many challenges, especially in terms of the accuracy of coupler positioning. Accurate coupler positioning is a key link to ensure the success and safety of the coupling operation. Most existing technologies rely on pure vision solutions and laser ranging methods, but both of these methods have their own limitations, resulting in the problem of insufficient coupler positioning accuracy.

[0019] In some technical solutions, the automatic coupling system mainly completes the detection and positioning of the coupler through visual recognition technology and lidar ranging. The visual recognition technology can obtain the image data of the target coupler through a camera or other imaging devices and analyze it through image processing algorithms. However, the visual solution is extremely vulnerable to environmental factors. For example, in rainy, foggy weather or a heavily smoky environment, the camera cannot clearly capture the image of the target coupler, resulting in a decrease in image recognition accuracy or even an inability to identify the position of the target coupler. Even worse, the camera is prone to misjudgment or missed judgment in low light or large light changes.

[0020] In addition, the laser ranging method relies on the lidar system to measure the distance to the target coupler. Although the laser ranging technology has high reliability under certain environmental conditions, the lidar usually can only provide the distance information of the target and cannot accurately determine whether the vehicle and the target coupler have been successfully docked. The single laser ranging method cannot effectively determine whether the coupling operation is completed. Therefore, although the laser ranging can provide certain distance information, its limitations make it difficult to meet the requirements of high-precision coupling operations.

[0021] Therefore, there are a series of pain points in the application of vision and lidar technologies in automatic coupling. First of all, the vision system is easily affected by the environment, and factors such as blurred images, insufficient light, and smoke interference will all lead to recognition failure or reduced accuracy. Secondly, the laser ranging method cannot comprehensively obtain the geometric information of the coupler, and can only provide the target distance data, lacking a comprehensive judgment on the shape and docking state of the target coupler. Therefore, these two technical solutions are difficult to ensure the precise execution of automatic coupling in various complex environments, especially in extreme weather conditions.

[0022] In summary, in the related technologies, there is a technical problem that the coupler positioning is not accurate enough during the automatic coupling of two rail vehicles.

[0023] As Figure 1 shown, this embodiment provides an automatic coupling method for a rail vehicle, and the method includes: Step S10: Control the first rail vehicle to travel at a preset first speed, and continuously collect the three-dimensional point cloud data in front of the first rail vehicle through a preset lidar during the travel; wherein, in front of the first rail vehicle refers to in front of the body coupler of the first rail vehicle; the lidar is arranged on the first rail vehicle and is on the same side as the body coupler.

[0024] In this embodiment, the execution subject of the method may be a control device or a control system of a first rail vehicle. Specifically, the control device may be an MCU (Micro Control Unit), an FPGA (Field Programmable Gate Array), an embedded controller, an industrial PC (IPC), a PLC (Programmable Logic Controller), or a single-chip microcomputer, etc.

[0025] In this embodiment, the first rail vehicle is a rail vehicle that needs to travel to perform coupling. It can be understood that in a specific scenario example, a first rail vehicle and a second rail vehicle are parked on a certain track. A body coupler and a lidar are provided on one side of the first rail vehicle. Correspondingly, a target coupler is provided on one side of the second rail vehicle. For example, a body coupler and a lidar are provided at the rear end position of the first rail vehicle, and a target coupler is provided at the front end of the second rail vehicle. Therefore, the body coupler and the target coupler can perform a coupling operation, and then the first rail vehicle and the second rail vehicle are connected.

[0026] Specifically, the first rail vehicle may be a complete locomotive, which can be located at the front end of the train and can have a power system and a control system or a control device, responsible for pulling and controlling the operation of the entire train. The second rail vehicle may be a complete carriage, which can be a passenger car or a freight car. The first rail vehicle may also be one of multiple carriages, which can be located at the front end or in the middle of the train and can be connected to other carriages. This carriage can also be equipped with a control system to coordinate the coupling operation. The second rail vehicle may be another carriage in the same train, waiting to be connected to the first rail vehicle. The first rail vehicle can be used as a power unit, which can be an electric locomotive or a diesel locomotive, providing traction and control functions. The second rail vehicle can be used as a trailer unit, which can be a non-powered carriage or a freight car, relying on the traction of the power unit.

[0027] Therefore, in this embodiment, the front of the first rail vehicle refers to the front of the body coupler of the first rail vehicle. That is, when the body coupler is provided at the rear end position of the first rail vehicle, the front of the first rail vehicle is also the front of the rear end. Correspondingly, when the body coupler is provided at the front end position of the first rail vehicle, the front of the first rail vehicle is also the front of the front end. In other words, the front of the first rail vehicle is the front in the traveling direction of the first rail vehicle, that is, the front of the body coupler (a second rail vehicle is parked at a position several distances in front of the front of the body coupler).

[0028] In this embodiment, the first rail vehicle and the second rail vehicle may be rail vehicles in an urban rail transit scenario. For example, urban subways or light rails.

[0029] In this embodiment, the first rail vehicle and the second rail vehicle may be rail vehicles in a long-distance railway transportation scenario. For example, they may be trains.

[0030] In this embodiment, the first rail vehicle and the second rail vehicle may be rail vehicles in an industrial rail system scenario. For example, they may be special locomotives in industrial parks or mining areas.

[0031] In this embodiment, the body coupler may be a Janney coupler. Correspondingly, the target coupler is a Janney coupler that can be adapted to the body coupler. These two couplers can perform coupling or uncoupling in an automatic or manual manner.

[0032] In this embodiment, the lidar may be disposed on the top of one side of the first rail vehicle for acquiring three-dimensional point cloud data of the front area on that side. One or more lidars may be provided.

[0033] In this embodiment, the lidar may be a mechanically rotating lidar.

[0034] In this embodiment, the lidar may be a solid-state lidar.

[0035] In this embodiment, the lidar may be a multi-beam lidar.

[0036] In this embodiment, the lidar may be a single-beam lidar.

[0037] In this embodiment, the lidar may be a long-range lidar.

[0038] It can be understood that the lidar may be communicatively connected to the control device or control system of the first rail vehicle. Specifically, the lidar may be fixed to the roof of the first rail vehicle or the rigid structure at the front of the vehicle body. The optical axis of the lidar is inclined downward to ensure that the scanning range covers the area where the coupler is located. The three-dimensional point cloud data collected by the lidar can be transmitted through Ethernet (UDP protocol) or a dedicated LVDS interface, and thus transmitted to the control device or control system of the first rail vehicle. And the control signal sent by the control device or control system of the first rail vehicle can be sent to the lidar or other modules (such as the power module of the rail vehicle) through the CAN bus.

[0039] It can also be understood that the lidar may also be pre-calibrated.

[0040] In this embodiment, the first speed is a pre-set speed, which can be the fastest driving speed during the process of the first rail vehicle performing automatic coupling. It can be understood that during the process of the first rail vehicle performing automatic coupling, the speed of the first rail vehicle can vary. For example, the speed of the first rail vehicle can be controlled according to the strategy of uniform speed approaching - contact deceleration - precise coupling. It can also be based on the relative distance between the first rail vehicle and the second rail vehicle to control the speed of the first rail vehicle. For example, the closer the distance, the smaller the speed of the first rail crane (for example, it can be dynamically adjusted through the PID control algorithm).

[0041] In this embodiment, the first speed can be 0.4 m / s, 0.5 m / s, or 0.6 m / s, etc. Specifically, the first speed is the highest initial driving speed pre-set by the rail vehicle during the automatic coupling process, used to quickly approach the target vehicle in the initial stage. The role of the first speed is that, on the one hand, it can shorten the approaching time as much as possible within the safety range, on the other hand, it can provide sufficient distance buffer for the subsequent precise alignment and deceleration stages, and on the other hand, it can be used as a reference parameter for speed hierarchical control to support the speed adjustment logic in different scenarios. For example, in a possible and specific embodiment, during the uniform speed approaching stage, it can travel at a uniform speed with the first speed (such as 0.5 m / s) to quickly reduce the distance from the target vehicle; during the contact deceleration stage, when the docking deviation (relative distance) reaches the preset threshold, it can switch to the second speed (such as 0.3 m / s) to enter the precise alignment mode, and in this stage, the speed can be fine-tuned in real time according to the docking deviation (such as the PID control algorithm). During the precise coupling stage, it can switch to the third speed when approaching contact to ensure low-speed contact and avoid mechanical shock.

[0042] Step S12: Based on the three-dimensional point cloud data, determine whether there is a target coupler; wherein, the target coupler is configured on one side of the second rail vehicle relative to the body coupler of the first rail vehicle.

[0043] In this embodiment, the target coupler is a coupler configured on the second rail vehicle, and this coupler needs to be connected to the body coupler. Specifically, the target coupler is fixedly installed at the end of the second rail vehicle, and its installation position is symmetrically arranged with the body coupler of the first rail vehicle. For example, if the body coupler of the first rail vehicle is located on the left side of the vehicle body, the target coupler is correspondingly installed on the right side of the second rail vehicle; the two can meet the coaxial alignment or the alignment conditions within the preset angular deviation range in space to ensure the feasibility of mechanical meshing.

[0044] In this embodiment, the point cloud data collected by the lidar can be used to reconstruct the three-dimensional geometric profile and spatial pose of the target coupler in real time.

[0045] In this embodiment, the step of determining whether there is a target coupler based on the three-dimensional point cloud data may include: First, perform feature extraction on the three-dimensional point cloud data to obtain feature data for characterizing the coupler features.

[0046] Specifically, the feature data for characterizing the coupler features may include: 1. The hook head surface feature. For example, it can be a curvature distribution histogram, a surface fitting parameter (quadratic surface coefficient), or a principal curvature direction (maximum / minimum curvature), etc. It can be understood that this hook head surface feature can be used to describe the geometric shape of the concave surface of the hook head.

[0047] 2. The hook tail plane feature. For example, it can be a plane equation parameter, a plane area, or a plane normal vector direction, etc. It can be understood that this hook tail plane feature can be used to describe the plane structure and spatial attitude of the hook tail.

[0048] 3. The spatial relationship feature. For example, it can be the hook head-hook tail distance, the perpendicular distance from the hook head center point to the hook tail plane, or the angle between the hook head normal vector and the hook tail normal vector, etc.

[0049] 4. The statistical feature. For example, it can be the point cloud density, the reflection intensity distribution (to distinguish materials), or the symmetry score (left-right symmetry of the hook head), etc. This statistical feature can be used to exclude foreign object interference (such as metal brackets, track reflections).

[0050] In this embodiment, a partition fitting algorithm based on curvature can be used to extract the hook head surface feature.

[0051] In this embodiment, the RANSAC (Random Sample Consensus) algorithm can be used to extract the hook tail plane feature.

[0052] In this embodiment, a geometric relationship analysis algorithm can be used to extract the spatial relationship feature.

[0053] Then, based on the feature data for characterizing the coupler features, determine whether there is a target coupler in front of the main coupler.

[0054] In this embodiment, the feature data for characterizing the coupler features can be matched with a preset feature template (which can store various coupler features for characterizing the target coupler) to determine whether there is a target coupler. It can be understood that when the preset similarity threshold is met, it is determined that there is a target coupler in front of the main coupler. Correspondingly, if the similarity threshold is not met, it is determined that there is no target coupler in front of the main coupler (the first rail vehicle and the second rail vehicle are still far apart).

[0055] In this embodiment, it is also possible to first use the feature data for characterizing the coupler features to further extract the point cloud of the ROI region (which can be the complete point cloud segment where the coupler head is located) from the three-dimensional point cloud data. Then, use the point cloud of this ROI region to match with a preset coupler point cloud template to determine whether there is a target coupler.

[0056] Step S14: When it is determined that there is a target coupler, calculate the docking deviation data between the target coupler and the body coupler; wherein, the docking deviation data at least includes the distance deviation data of the docking.

[0057] In this embodiment, the docking deviation data may further include the docking angle deviation data. Specifically, the docking deviation data can be expressed as a set of quantization indexes characterizing the comprehensive deviation degree of the spatial postures of the two couplers. The distance deviation data of the docking can be used to describe the straight-line distance between the center points of the two couplers in a specified coordinate system. The docking angle deviation data can be used to describe the spatial angle difference between the normal vector directions of the two couplers.

[0058] More specifically, the distance deviation data of the docking can be used to quantify the relative position deviation of the center points of the two couplers in three-dimensional space, which directly affects the approaching speed and path planning. The distance deviation data of the docking can be the three-dimensional Euclidean distance (fully characterizing the spatial deviation), or the two-dimensional projection distance (that is, the Z-axis is fixed and only horizontal alignment on the plane is required), or the one-dimensional distance data (that is, both the Z-axis and the Y-axis are fixed and only the distance of the X-axis needs to be determined. It can be understood that the track is the X-axis, the axis perpendicular to the ground is the Z-axis, and on the ground, the axis perpendicular to the X-axis is the Y-axis). The docking angle deviation data can be used to quantify the alignment error of the normal vectors of the two couplers to ensure no lateral stress during mechanical meshing. Similarly, this docking angle deviation data can be the spatial angle between the normal vectors (three-dimensional angle), or the yaw angle in the projection plane (two-dimensional simplification), or the one-dimensional included angle (one-dimensional simplification).

[0059] In a specific implementation scheme, the step S14 may include: The step of calculating the docking deviation data between the target coupler and the body coupler when it is determined that there is a target coupler includes: Step S142: Based on the latest three-dimensional point cloud data, extract the target three-dimensional point cloud data; wherein, the target three-dimensional point cloud data is the three-dimensional point cloud data representing the region where the target coupler is located; Step S144: For the target three-dimensional point cloud data, execute a preset surface fitting algorithm to generate a fitted quadratic surface model; Step S146: Based on the quadratic surface model, calculate the center point coordinates and the normal vector direction of the target coupler; Step S148: Based on the pre-calibrated center point coordinates and normal vector direction of the main body coupler and the center point coordinates and normal vector direction of the target coupler, the distance deviation data and the docking angle deviation data are calculated.

[0060] Step S16: Based on the distance deviation data and a preset speed control strategy, the travel speed of the first rail vehicle is controlled so that the first rail vehicle approaches the second rail vehicle at a speed not higher than and not equal to the first speed.

[0061] In this embodiment, the preset speed control strategy may be a speed-graded control strategy. For example, a plurality of distance intervals may be pre-set based on the distance deviation data, and different distance intervals correspond to different speeds. Within a certain distance interval, the first rail vehicle is controlled to maintain the speed corresponding to the distance interval to perform uniform speed travel. The closer the first rail vehicle is to the second rail vehicle (that is, the smaller the distance deviation data is), the smaller the speed is.

[0062] In this embodiment, the preset speed control strategy can be a speed control strategy in which the driving speed decreases linearly. More specifically, the entire approach distance can be divided into multiple intervals (such as 0-0.3 meters, 0.3-1 meters, and more than 1 meter). When the distance is greater than 1 meter, the speed is constant at 0.4m / s. When the distance is 0.3-1 meters, the speed decreases linearly from 0.4m / s to 0.1m / s. When the distance is less than 0.3 meters, the speed further decreases linearly to 0.05m / s. And a gradual algorithm is used at the interval boundary to avoid sudden changes in speed. In a specific scenario example, the vehicle approaches from 2 meters away with an initial speed of 0.4m / s; when the distance is reduced to 0.8 meters, the speed is proportionally reduced to 0.25m / s; when the distance is 0.2 meters, it enters the final fine-tuning stage and the speed is reduced to 0.05m / s to ensure impact-free contact.

[0063] In this embodiment, the preset speed control strategy may be a speed control strategy in which the driving speed is non-linearly reduced.

[0064] In this implementation, the preset speed control strategy may also be a control strategy based on PID feedback.

[0065] Specifically, the distance deviation data can be used as input to calculate the speed adjustment in real time through the proportional (P), integral (I), and differential (D) controllers.

[0066] In the formula, can be expressed as the adjusted speed, It can be expressed as the real-time deviation between the target value and the actual value (distance deviation data). can be expressed as the integral gain coefficient, can be expressed as the integral term, that is, the cumulative sum of errors from the initial time to the current time t, can be expressed as the differential gain coefficient, can be expressed as the rate of change of the error with respect to time, i.e., the derivative of the error.

[0067] In this embodiment, the preset speed control strategy can also be a speed control strategy based on fuzzy logic control. For example, the fuzzy sets of distance deviation (such as "far", "medium", "near") and the fuzzy rules of speed output can be defined, and the non-linear mapping is realized through the membership function. More specifically, the precise distance deviation (such as 1.5 meters, 0.6 meters) can be converted into fuzzy language values such as "far", "medium", "near". For example: when the distance > 1 meter, the degree of membership in "far" is 80%; when the distance is 0.3 - 1 meter, the degree of membership in "medium" is 60%; when the distance < 0.3 meter, the degree of membership in "near" is 90%. Fuzzy rules can be defined. For example, if the distance is "far", then "high speed" is output; if the distance is "medium" and the angle deviation is "small", then "medium speed" is output; if the distance is "near" or the angle deviation is "large", then "low speed" is output. Finally, the fuzzy output is converted into an exact speed value (such as "high speed" corresponding to 0.4 m / s, "low speed" corresponding to 0.05 m / s). In a specific scenario example, when the vehicle is 2 meters away from the target, the fuzzy system determines it as "far" and approaches at a constant speed of 0.4 m / s; when the distance drops to 0.5 meters, the system comprehensively judges the angle deviation and determines it as "medium speed", and the speed is adjusted to 0.2 m / s; when the distance is 0.2 meters, "low speed" is triggered and the vehicle enters millimeter-level fine movement.

[0068] In this embodiment, the preset speed control strategy can also be a speed control strategy based on machine learning or deep learning. More specifically, historical operation data can be collected in advance, including distance deviation, angle deviation, speed command, and hooking result (success / failure). Then, the historical operation data is used to train the neural network, and the neural network (such as a fully connected network or LSTM) is used to learn the mapping relationship between the input (distance, angle) and the output (speed), and the model is optimized through supervised learning, with the goal of minimizing the difference between the predicted speed and the actual operation speed. In a specific scenario example, in a dusty environment, the model learns from historical data that "the speed needs to be reduced in advance when the dust concentration is high". When it detects that the current dust concentration exceeds the standard, even if the distance is 1 meter, the speed is actively reduced from 0.4 m / s to 0.3 m / s to avoid the risk of misjudgment.

[0069] In this embodiment, the preset speed control strategy can also be a speed control strategy based on reinforcement learning (RL). For example, a speed control strategy network can be trained through a deep reinforcement learning model, and the reward function is designed to balance fast approaching and low impact force.

[0070] Therefore, in this embodiment, the process of making the first rail vehicle approach the second rail vehicle at a speed not higher than and not equal to the first speed can be expressed as follows: during the process of the first rail vehicle driving towards the second rail vehicle, any one of the above speed control strategies is adopted to reduce the first speed. For example, it can be reduced in a linear manner, or in a non-linear manner, or it can also be reduced by means of model prediction, so that the first rail vehicle approaches the second rail vehicle at a speed not higher than and not equal to the first speed.

[0071] Step S18: When it is determined that the preset coupling distance has been reached, execute the closing of the coupling.

[0072] In this embodiment, the preset coupling distance can be a relatively short preset distance. For example, it can be 5 cm, or 4 cm or 6 cm, etc. It can be understood that on the one hand, there is a certain amount of noise and delay in the point cloud data collected by the lidar during the vehicle movement, and the typical ranging error is about ±1 cm. Therefore, the preset distance is generally selected in the range of 4 - 6 cm, with 5 cm being the preferred value. On the one hand, it can ensure sufficient safety margin, and on the other hand, it can leave enough response time before mechanical closing to avoid impact or misalignment. On the other hand, the coupler head needs a certain buffer stroke to insert into the coupler mouth. If it is too short, the closing action will not be completed in time, and if it is too long, the risk of secondary collision will increase. A distance of about 5 cm can ensure that the coupler head has basically entered the coupler mouth and overlaps with the alignment range of the locking pin in most Janney coupler structures.

[0073] This embodiment precisely controls the automatic coupling process of the rail vehicle by combining the lidar and the three-dimensional point cloud data, and has significant beneficial effects. First, by continuously collecting three-dimensional point cloud data through the lidar, the position of the target coupler can be monitored in real time, without being interfered by environmental factors such as light, rain and snow, overcoming the recognition deficiencies of traditional vision systems in complex weather. Second, by calculating the distance deviation data between the couplers and adjusting the driving speed according to the preset speed control strategy, it is ensured that the first rail vehicle can accurately approach the target coupler at an appropriate speed, reducing the risk of collision or misalignment. In addition, after it is determined that the preset coupling distance has been reached, the coupling closing operation is automatically executed, ensuring the automation and high efficiency of the coupling process, reducing the need for manual intervention, and improving the operation safety and reliability. Therefore, this embodiment has high precision and adaptability in the automatic coupling of rail transit, can improve the operation efficiency and reduce the safety hazards caused by misoperation.

[0074] In some embodiments, the step of determining whether there is a target coupler based on the three-dimensional point cloud data includes: Step S122: Perform feature extraction on the three-dimensional point cloud data to obtain feature data for characterizing the coupler features.

[0075] Specifically, the feature data for characterizing the coupler features may include: 1. The hook head surface feature. For example, it can be a curvature distribution histogram, it can be surface fitting parameters (quadratic surface coefficients), or it can be the principal curvature direction (maximum / minimum curvature), etc. It can be understood that this hook head surface feature can be used to describe the geometric shape of the concave surface of the hook head.

[0076] 2. The hook tail plane feature. For example, it can be plane equation parameters, it can be the plane area, or it can be the plane normal vector direction, etc. It can be understood that this hook tail plane feature can be used to describe the plane structure and spatial attitude of the hook tail.

[0077] 3. The spatial relationship feature. For example, it can be the hook head - hook tail spacing, it can be the perpendicular distance from the center point of the hook head to the hook tail plane, or it can be the angle between the hook head normal vector and the hook tail normal vector, etc.

[0078] 4. The statistical feature. For example, it can be the point cloud density, it can be the reflection intensity distribution (to distinguish materials), or it can be the symmetry score (the left - right symmetry of the hook head), etc. This statistical feature can be used to exclude foreign object interference (such as metal brackets, track reflections).

[0079] In this embodiment, a curvature - based partition fitting algorithm can be used to extract the hook head surface feature.

[0080] In this embodiment, the RANSAC (Random Sample Consensus) algorithm can be used to extract the hook tail plane feature.

[0081] In this embodiment, a geometric relationship analysis algorithm can be used to extract the spatial relationship feature.

[0082] In some embodiments, PCA analysis can also be performed on the neighborhood of each point to calculate the point cloud normal vector and the principal curvature direction; on this basis, local descriptors such as Fast Point Feature Histograms (FPFH) or Signature of Histograms of Orientations (SHOT) can be further constructed to encode the surface shape and normal vector distribution for distinguishing the coupler surface from the surrounding metal or debris.

[0083] In some embodiments, an end-to-end feature extraction algorithm based on deep learning can also be used to perform feature extraction to obtain feature data for characterizing the coupler features. For example, networks such as PointNet / PointNet++, DGCNN, and PointCNN can be used to directly learn high-dimensional feature vectors from the original or downsampled point cloud input. The above-mentioned networks can be pre-trained through classification or segmentation tasks and fine-tuned in the coupler scenario to obtain discriminative depth features for components such as the coupler head, coupler tail, and locking pin.

[0084] Step S124: Based on the feature data for characterizing the coupler features, determine whether there is a target coupler in front of the body coupler.

[0085] In this embodiment, the feature data for characterizing the coupler features can be matched with a preset feature template (which can store various coupler features for characterizing the target coupler) to determine whether there is a target coupler. It can be understood that when the preset similarity threshold is met, it is determined that there is a target coupler in front of the body coupler. Correspondingly, if the similarity threshold is not met, it is determined that there is no target coupler in front of the body coupler (the first rail vehicle and the second rail vehicle are still far apart).

[0086] Specifically, the preset feature template can be pre-stored with: Geometric parameters: coefficients of the coupler head surface equation, coupler tail plane equation, center point coordinates, normal vector direction; Statistical parameters: curvature distribution histogram, average reflection intensity, symmetry score; Spatial relationship: standard distance between the coupler head and the coupler tail, normal vector angle range.

[0087] When performing the matching, rough matching can be performed first, and then fine matching (ICP registration). More specifically, the cosine similarity between the real-time coupler head curvature histogram and the template can be calculated, and those with too low cosine similarity can be directly excluded (through a preset similarity threshold, for example, 85%). In this way, more than 90% of non-target objects (such as metal brackets beside the track) can be quickly excluded. Then, the real-time coupler head point cloud is aligned with the template point cloud, the registration error is calculated (root mean square error < 5mm), and it is verified whether the distance between the coupler head and the coupler tail and the normal vector angle are within the standard range.

[0088] In this embodiment, the feature data for characterizing the coupler features can also be used to further extract the point cloud of the ROI region (which can be the complete point cloud segment where the coupler head is located) from the three-dimensional point cloud data. Then, the point cloud of this ROI region is matched with the preset coupler point cloud template to determine whether there is a target coupler.

[0089] Specifically, first, a point cloud template for the coupler can be pre-built. For example, in a laboratory environment, a high-resolution laser radar (such as 128 lines) can be used to scan a standard coupler to obtain a noise-free point cloud, which can also include coupler models with different degrees of wear and installation angles (±5° tilt). Then, the hook head area is located based on the feature data. The curvature value of each point in the point cloud can be calculated first, and the high curvature area (curvature>0.05) can be screened to preliminarily locate the candidate area of ​​the hook head. Then, the high curvature points are clustered in Euclidean clustering (the clustering radius can be 10 cm), and the largest cluster is retained as the core area of ​​the hook head. Then, the ROI area is delineated. The core area of ​​the hook head can be used as the center, and several centimeters (can be 20 cm) can be extended along the track extension direction (X axis), several centimeters (can be 30 cm) in the vertical track direction (Y axis), and several centimeters (can be 20 cm) in the height direction (Z axis) to form a three-dimensional ROI area. A pass-through filter can be further applied to accurately crop the ROI point cloud to exclude irrelevant objects (such as tracks and gravel). You can further count the outlier noise points to remove the discrete noise points in the ROI. Finally, you can use the ROI point cloud to match the template. You can also perform a rough match first. For example, you can calculate the curvature distribution histogram of the ROI point cloud, calculate the cosine similarity with the curvature histogram of all the hooks in the template library, and select candidate templates with a similarity of > 0.8. Then check whether the hook head-hook tail spacing is within the standard range (500±10mm) and exclude templates that are obviously inconsistent. After performing the rough match, perform a fine match. For example, you can use the iterative closest point (ICP) algorithm to align the ROI point cloud with the candidate template point cloud and optimize the rigid body transformation matrix (translation + rotation). After alignment, you can calculate the root mean square error (RMSE) between the two point clouds, and check whether the angle between the two hook head normal vectors is less than the angle threshold (for example, it can be 1 degree) to ensure posture alignment.

[0090] By extracting features from the three-dimensional point cloud data in step S122 and judging the existence of the target coupler based on the extracted coupler feature data in step S124, this embodiment can quickly and accurately identify the object to be hooked in a complex environment. Compared with the rough detection directly based on the original point cloud or two-dimensional image, this method uses multi-dimensional geometric features such as the hook head surface, the hook tail plane and the spatial relationship, which can not only effectively filter environmental clutter (such as interference from track facilities, foreign body reflections, etc.), but also reduce the risk of misjudgment caused by occlusion or partial point cloud missing; at the same time, through feature template matching rather than simple threshold judgment, higher recognition accuracy and faster response speed are achieved, providing reliable input for subsequent deviation calculation and speed control, and significantly improving the stability and intelligence level of the system.

[0091] In some embodiments, the docking deviation data further includes: docking angle deviation data; The step of calculating the docking deviation data between the target coupler and the body coupler when it is determined that the target coupler exists includes: Step S142: Extract target three-dimensional point cloud data based on the latest three-dimensional point cloud data; wherein, the target three-dimensional point cloud data is the three-dimensional point cloud data representing the area where the target coupler is located.

[0092] In this embodiment, the corresponding target three-dimensional point cloud data in the latest three-dimensional point cloud data can be extracted based on the feature data of the latest three-dimensional point cloud data.

[0093] Specifically, first, for each point in the latest three-dimensional point cloud data, calculate the corresponding local feature descriptor (such as curvature, normal vector, FPFH / SHOT, etc.). Then, match the local feature descriptor of each point with a preset feature template to obtain a similarity score. Next, the high-score points that best match the feature template in the latest three-dimensional point cloud data can be selected as "seed points". To prevent misselection, it can also be required that the seed points be spatially aggregated - that is, only when a certain number of high-score points appear in a certain area, will this area be used as the starting point for subsequent extraction. Then, perform a region-growing algorithm based on features. For example, each seed point can be used as the center to search for adjacent points along the spatial neighborhood (for example, with a radius r = 5 cm); for each adjacent point, if the similarity between its features and the feature template meets the threshold, it is also marked as a "coupler candidate point" and continues to grow outwards; points that do not meet the conditions are not included, and at the same time, prevent crossing to other objects. Then, perform Euclidean Clustering on all the marked candidate points, and divide them into several clusters. Clusters that are significantly too large, too small, or have a position deviation that is too far can be removed according to the size of the cluster, geometric shape (range of length, width, and height), and the distance between the cluster center and the expected coupler position. Finally, select the cluster that best matches the coupler features (it can be the one with the largest number of points or the highest average similarity) from the remaining clusters, and output all the point sets corresponding to this cluster as the "target three-dimensional point cloud data" for the next surface fitting (S144) and deviation calculation (S148).

[0094] In this embodiment, the target three-dimensional point cloud data can also be extracted from the latest three-dimensional point cloud data based on prior knowledge.

[0095] Specifically, this prior knowledge can be a constraint on the spatial range, such as: X-axis (track direction): 0.5 m to 5 m in front of the body coupler (to avoid scanning irrelevant areas that are too far away); Y-axis (perpendicular to the track direction): ±0.5 m (covering the lateral swing range of the coupler); Z-axis (height direction): 0.2 m to 0.8 m above the track surface (the range of the coupler installation height).

[0096] The pass-through filtering can be performed again on the point cloud data that meets the spatial range constraints, and the point cloud within the ROI is retained.

[0097] Step S144: For the target three-dimensional point cloud data, perform a preset surface fitting algorithm to generate a fitted quadratic surface model.

[0098] Specifically, this step S144 can further include: First, before performing the preset surface fitting algorithm, the statistical outlier removal (SOR) algorithm can be performed again to remove noise points. The reflection intensity correction and data normalization can also be performed (for example, the point cloud coordinate system can be converted to a local coordinate system with the center of the target coupler as the origin to eliminate the influence of global position offset). Then, use the least squares method to solve the preset quadratic surface equation to obtain the fitted quadratic surface model.

[0099] More specifically, the preset quadratic surface equation can be: In the formula, , and can be parameters for controlling the curvature and direction of the surface, and can be parameters for controlling the tilt of the surface, is the translation term, which is used to adjust the reference height of the surface. It can be understood that - are the model parameters to be solved.

[0100] The following error function can be constructed in advance to solve the parameters to be solved in the above quadratic surface equation.

[0101] In the formula, is the weight coefficient, and higher weights can be assigned to the points in the high-curvature region, is the number of point clouds.

[0102] Through this error function, the sum of the squares of the vertical distances from each data point to the surface can be defined as the optimization objective.

[0103] In this embodiment, the error function can be converted into a linear equation form for solution. For example: In the formula, represents the design matrix, and each row corresponds to the of a point, is the diagonal weight matrix, is the parameter to be solved , .

[0104] The linear equations can be solved by QR decomposition or SVD to obtain the optimal parameters .

[0105] Step S146: Calculate the center point coordinates and normal vector direction of the target coupler based on the quadratic surface model

[0106] Specifically, the method of surface vertex positioning can be used to calculate the center point coordinates of the target coupler. Specifically, by utilizing the inherent "vertex" characteristic of the quadratic surface, the overall shape of the surface model is scanned to automatically find the point where the inclination of the surface is exactly zero in all directions, that is, the position where the surface is the "flattest" and most "bulging". This position is identified as the geometric center of the coupler head

[0107] The method of point cloud centroid projection can also be used to calculate the center point coordinates of the target coupler. Specifically, first calculate the overall average position (centroid) of the target point cloud in step S142, and then project it onto the surface along the "shortest path" of the fitted surface to find the surface position closest to the centroid. This projection point is both closest to the center of the original point cloud and lies on the surface, and is regarded as the center of the coupler head

[0108] More specifically, first, the target point cloud extracted in step S142 is averaged arithmetically to obtain the centroid Then, project the centroid onto the fitted surface, and the minimum distance projection can be used In the formula, is the gradient vector calculated at the centroid position, is the implicit function value

[0109] Then, the projection point can be obtained through the Newton iteration method as the approximate "center point coordinates" of the coupler head

[0110] Finally, the gradient can be calculated at the projection point and normalized to obtain the surface normal vector direction

[0111] Step S148: Calculate the distance deviation data and docking angle deviation data based on the pre-calibrated center point coordinates and normal vector direction of the body coupler and the center point coordinates and normal vector direction of the target coupler

[0112] In this embodiment, the central points of the body coupler and the target coupler can be directly regarded as two endpoints in space, and the shortest straight-line distance between them can be measured. For example, it can be the three-dimensional Euclidean distance between the central points of the two couplers, which directly represents the spatial position deviation.

[0113] The docking angle deviation data can be the spatial angle between the normal vectors of the two couplers, which directly represents the attitude alignment deviation. For example, the dot product of two unit vectors can be calculated, and then the inverse cosine function is taken on the result to obtain a radian value, and finally it is converted into an angle value.

[0114] By introducing quadratic surface fitting and normal vector calculation in steps S142–S148, this embodiment can not only obtain the three-dimensional distance deviation between the target coupler and the body coupler, but also accurately quantify the angle deviation between the normal vectors of the two, so as to realize the dual alignment control of attitude and position. Compared with the method that only relies on distance information, the angle deviation data can avoid lateral stress or meshing jamming caused by slight inclination of the coupler head; at the same time, the smooth model obtained by quadratic surface fitting can suppress the interference of point cloud noise and improve the stability of normal vector estimation. Overall, when dynamically planning the vehicle travel path and speed grading, this scheme can adjust the heading in real time according to the angle deviation and fine-tune the wheel set steering, realizing centimeter-level and degree-level precise alignment, greatly reducing the risk of hooking impact or misalignment, and improving the hooking success rate and system reliability.

[0115] In some embodiments, the step of controlling the driving speed of the first rail vehicle based on the distance deviation data and a preset speed control strategy includes: Step S162: When it is determined that there is a target coupler, switch the first speed to the second speed, and control the first rail vehicle to travel at a constant speed at the second speed; wherein, the second speed is lower than the first speed.

[0116] In this embodiment, the first speed is a preset speed, which can be the fastest driving speed during the process of the first rail vehicle performing automatic hooking. It can be understood that during the process of the first rail vehicle performing automatic hooking, the speed of the first rail vehicle can change. For example, the speed of the first rail vehicle can be controlled according to the strategy of approaching at a constant speed - decelerating when contacting - precisely hooking. It can also be based on the relative distance between the first rail vehicle and the second rail vehicle to control the speed of the first rail vehicle. For example, the closer the distance, the lower the speed of the first rail crane (for example, it can be dynamically adjusted through a PID control algorithm). In this embodiment, the first speed can be 0.4 m / s, 0.5 m / s or 0.6 m / s, etc.

[0117] In this embodiment, the second speed can be 0.2 m / s, 0.1 m / s, etc. It can be understood that, on the one hand, compared with the first speed, the second speed significantly reduces the kinetic energy of the vehicle, enabling the first rail vehicle to maintain a more stable uniform speed when initially approaching the target coupler, reducing overshoot or jitter caused by inertia, and ensuring the quality of point cloud acquisition and processing. On the other hand, although the speed is reduced, it is still sufficient to support the vehicle to approach the target within a short time, leaving enough distance buffer for the next lower-speed fine-tuning; at the same time, in the transition interval from a few meters to dozens of centimeters away from the target coupler, the second speed can quickly respond to control commands to achieve precise attitude correction. Moreover, at this speed gear, both the lidar point cloud sampling and algorithm processing can maintain high frame rate and stability, avoiding point cloud blur or omission caused by too fast movement, and providing a reliable data basis for subsequent precise alignment.

[0118] Step S164: Determine whether the distance deviation data reaches a preset speed switching threshold.

[0119] In this embodiment, the preset speed switching threshold can be 0.3 m, 0.5 m, 0.2 m, etc.

[0120] In this embodiment, the speed switching threshold is a preset critical distance value used to trigger the phased adjustment of the vehicle speed. When the real-time distance deviation between the couplers reaches this threshold, the driving speed will be automatically reduced to improve the docking accuracy and avoid the risk of collision.

[0121] Step S166: When the preset speed switching threshold is reached, switch the second speed to the third speed and control the first rail vehicle to travel at the third speed; wherein, the third speed is lower than the second speed.

[0122] In this embodiment, the third speed can be 0.05 m / s, 0.06 m / s, 0.04 m / s, etc.

[0123] In this embodiment, the third speed is the ultra-low speed used by the first rail vehicle in the approaching final contact stage, which can move in millimeters and is used to achieve precise alignment and eliminate inertial impact.

[0124] In this embodiment, switching the second speed to the third speed and controlling the first rail vehicle to travel at the third speed can be to control the first rail vehicle to travel at a uniform speed at the third speed.

[0125] In this embodiment, switching the second speed to the third speed and controlling the first rail vehicle to travel at the third speed can also be to continuously reduce the driving speed on the basis of the third speed for traveling (for example, fine-tuning the speed through a PID control algorithm).

[0126] In this embodiment, after detecting the target coupler, the two-stage speed switching embodiment first switches from a higher speed to a medium speed and approaches the target smoothly; when the distance deviation reaches a preset threshold, it switches to the low-speed stage to ensure a gentle touch in the final docking stage. Through the hierarchical deceleration strategy of first fast and then slow, it not only shortens the overall approach time and improves the operation efficiency, but also provides sufficient buffering for accurate alignment and closing actions, greatly reducing the impact force and misalignment risk, ensuring high precision and high coupler connection success rate.

[0127] In some embodiments, based on the distance deviation data and a preset speed control strategy, the traveling speed of the first rail vehicle is controlled such that before the first rail vehicle approaches the second rail vehicle at a speed not higher than and not equal to the first speed, and in the case where the coupler is a Janney coupler, the method further includes: Sending a control signal to the power device of the body coupler to open the body coupler so that the body coupler is in a state to be connected; In this embodiment, the power device of the body coupler may be an electric actuator. For example, it can be driven by a DC motor or a servo motor, and the rotational motion is converted into a linear stroke through a reduction gear, a worm and worm gear or a ball screw to push and pull the coupler. It can also be a hydraulic cylinder system. For example, it can be composed of a hydraulic pump, valve components and a hydraulic cylinder, and high-pressure hydraulic fluid is used to push the piston to generate a large thrust. It can also be an electro-hydraulic servo actuator and so on.

[0128] In this embodiment, the opening of the body coupler may be to push the locking pin assembly through the above-mentioned power device, so that the locking pin exits (lifts) from the pin hole, releasing the fixed constraint on the coupler jaw. At this time, the coupler jaw can move freely in the subsequent steps. After the locking pin is disengaged, the power device continues to act on the rotating arm (lever or link) of the coupler jaw, and rotates the coupler jaw from the closed position counterclockwise (or clockwise, depending on the mechanism design) to a predetermined "open" angle.

[0129] The step of performing the coupler closing in the case of determining that the preset coupler connection distance is reached includes: Step S182: In the case of determining that the preset coupler connection distance is reached, controlling the first rail vehicle and the second rail vehicle to touch at a speed not higher than the third speed.

[0130] In this embodiment, the controlling the first rail vehicle and the second rail vehicle to touch at a speed not higher than the third speed may be directly controlling the first rail vehicle to travel at the third speed, or starting from the third speed and continuously reducing the speed (the closer the distance, the smaller the speed), so as to control the first rail vehicle and the second rail vehicle to touch.

[0131] Step S184: After detecting that a touch occurs, collect three-dimensional point cloud data of the hook area through a preset lidar.

[0132] In this embodiment, the detection of the occurrence of a touch can be judged by a preset force sensor to determine whether a touch occurs. For example, a force sensor can be preset on one side of the body coupler. It can also be dynamically analyzed through a preset inertial measurement unit (IMU).

[0133] Step S186: For the three-dimensional point cloud data of the hook area, extract the point cloud voxel block of the locking pin area of the Janney coupler.

[0134] In a specific implementation, it can include: after the hook is closed, first preprocess the high-density hook area point cloud collected by the lidar to remove environmental noise and unify the resolution. Specifically, first use the statistical outlier removal (SOR) algorithm to remove the outlier points with abnormal density, and then remove the isolated noise with very few surrounding points through radius filtering. Then, apply voxel grid filtering to the remaining point cloud to unify the resolution into small cubic blocks with a side length of about 2 mm per voxel.

[0135] After the preprocessing is completed, combined with the geometric calibration information of the Janney coupler, a candidate area for the locking pin to be extracted is cropped in the three-dimensional coordinate system at a certain distance (which can be in the range of 50 - 80 mm) below the center of the coupler head and within a horizontal radius of about 20 mm. In this area, the high-curvature point set is quickly locked through curvature analysis (the points with a curvature exceeding about 0.05 correspond to the surface of the cylindrical locking pin), and then the RANSAC cylindrical fitting algorithm is used. With the preset locking pin radius (which can be 25 - 40 mm) and height (which can be 80 - 120 mm) as constraints, the optimal cylindrical model is determined after multiple iterations. Finally, a cluster of voxels that best fits this cylindrical model is regarded as the locking pin area, and it is reorganized with a voxel size of 1 cm³, retaining the representative point with the maximum reflection intensity in each voxel, and outputting a voxel block of about 20×20×20, which completely covers the geometric information of the locking pin.

[0136] Step S188: Based on the extracted point cloud voxel block of the locking pin area and the preset locking pin landing model, perform a primary judgment on whether the hooking is successful.

[0137] In this embodiment, three reference voxel models of the locking pin landing states - "fully landed", "half landed", and "not landed" - can be stored in the template library in advance. Each model is generated from high-precision point clouds and has an FPFH (Fast Point Feature Histogram) descriptor to characterize local geometric features.

[0138] For the real-time locking pin voxel block obtained in step S186, its FPFH features can be extracted first, and the nearest neighbor search can be performed in the template library to calculate the average cosine similarity of all feature vectors. When the similarity ≥ similarity threshold (which can be 0.85), it indicates that the real-time voxel is highly consistent with a certain template in geometric distribution, and it can be initially determined as the corresponding state.

[0139] In a specific implementation, to enhance the accuracy, the candidate voxel block with a high score can be further subjected to ICP (Iterative Closest Point) registration. Through rigid body transformation, the real-time voxel is closely aligned with the template, and it is checked whether the root mean square error after registration ≤ 2mm. If both the ICP error and the FPFH match meet the standards, further multi-level verification is performed on geometric and optical features such as the angle between the locking pin axis and the pin hole axis (which needs to be less than 0.5°), the distance between the bottom of the locking pin and the bottom surface of the pin hole (which needs to be less than 1mm), and the average reflection intensity of the locking pin area (which needs to be greater than 80). If all are satisfied, "the first hook-up is successful" is output; otherwise, exception handling is triggered, such as retrying the docking or alarming to remind manual intervention.

[0140] Step S1810: In the case where it is initially determined that the hook-up is successful, execute a preset final judgment strategy to finally determine whether the hook-up is successful.

[0141] In this implementation, the main body coupler is automatically opened to the waiting state at a very low speed when approaching, and when touching, it contacts smoothly at an extremely low speed. Then, the high-density three-dimensional point cloud scanning of the hook-up area is carried out by using a lidar, and after accurately extracting the locking pin voxel, the initial locking pin landing judgment is completed. Finally, combined with reverse verification or other final judgment strategies for final confirmation. Thus, a closed-loop control from open hook-up, low-speed docking, real-time point cloud detection to double verification is formed, which not only ensures the smooth and impact-free hook-up action, but also can sense and correct the problem of the locking pin not landing or foreign object interference in the first time, greatly improving the hook-up success rate and system reliability, minimizing mechanical damage and safety risks, and realizing true unmanned and all-weather automatic hook-up.

[0142] In some implementations, the step of performing the initial judgment on whether the hook-up is successful based on the extracted point cloud voxel block of the locking pin area and the preset locking pin landing model includes: Step S1882: Match and compare the point cloud voxel block of the locking pin area with the pre-stored reference point cloud spatial distribution in the locking pin landing model to obtain the spatial alignment relationship between the two.

[0143] In this embodiment, first, two sets of voxels are roughly aligned using a coarse registration method (such as based on voxel centers or known hook postures). Then, a fine registration algorithm (Iterative Closest Point ICP) is run to continuously adjust the translation and rotation parameters in the three-dimensional space so that the real-time voxel block and the model voxel block achieve the best overlap in terms of overall shape. Finally, a set of rigid body transformation matrices is output, which is the "spatial alignment relationship" and can accurately describe how to map the real-time point cloud voxels to the model coordinate system.

[0144] Step S1884: Based on the spatial alignment relationship, extract the point cloud cluster corresponding to the locking pin end point from the point cloud voxel block, and calculate the center coordinates of the point cloud cluster.

[0145] In this embodiment, first, a small cluster of points (endpoint cluster) representing the locking pin end can be identified and separated in the transformed voxels because this cluster is the closest in space to the position of the locking pin end in the template. Then, clustering analysis is performed on this point cluster to remove surrounding miscellaneous points and only retain the densest endpoint cluster. Finally, the average position of all points in this cluster is calculated, which is the "center coordinate of the locking pin end point" and is used for subsequent tolerance judgment.

[0146] Step S1886: Determine whether the center coordinates of the point cloud cluster fall within the three-dimensional tolerance range of the calibrated hook pin hole area in the locking pin positioning model, and determine whether the absolute value of the coordinate deviation in the vertical direction is within a preset vertical deviation range.

[0147] In this embodiment, determining whether the center coordinates of the point cloud cluster fall within the three-dimensional tolerance range of the calibrated hook pin hole area in the locking pin positioning model can be to check the deviation distance of the endpoint center in the orifice plane (horizontal XY plane) relative to the center of the template hole, and it is required to be within a preset horizontal tolerance radius (such as ±5 mm). Determining whether the absolute value of the coordinate deviation in the vertical direction is within a preset vertical deviation range can be to compare the height difference of the endpoint center in the vertical direction (Z-axis), and it is required to fall within the depth tolerance range of the template hole (such as ±3 mm).

[0148] Step S1888: In the case where both are satisfied, it is determined that the locking pin has been correctly positioned, and it is initially determined that the hooking is successful.

[0149] In this embodiment, by precisely registering the real-time extracted locking pin voxel blocks with the pre-stored landing model, and then accurately locating the endpoints of the locking pin and calculating its spatial center from the aligned data, it is possible to determine whether the locking pin has truly fallen into the pin hole with high-resolution three-dimensional tolerance. This method avoids misjudgments caused by simply relying on overall shape matching. Through double tolerance verification of the locking pin endpoints in the horizontal and vertical directions, a "fine-grained" initial determination of the hook state is achieved, significantly improving the accuracy and reliability of the locking pin landing detection, thereby significantly reducing the repetitive operations and safety risks caused by misjudgments.

[0150] In some embodiments, the step of, in the case of initially determining that the hook is successful, executing a preset final judgment strategy to determine whether the hook is successful includes: Step S18102: Control the first rail vehicle to move backward a preset distance along the driving direction, and during the backward movement, continuously collect three-dimensional point cloud data of the locking pin area through the lidar.

[0151] In this embodiment, after initially determining that the locking pin has landed, the control device can automatically issue a control instruction for slight backward movement, causing the first rail vehicle to move a small preset distance (which can be from a few centimeters to a few tenths of a meter) in the opposite direction of the original driving direction. During this process, the on-vehicle lidar continuously collects point clouds of the locking pin area to ensure that multiple frames of continuous three-dimensional point cloud data can be obtained regardless of how the vehicle moves slightly, for subsequent stability analysis. This moving distance and speed are strictly controlled to prevent secondary disengagement or data blurring caused by excessive or too fast displacement.

[0152] Step S18104: Based on the three-dimensional point cloud data collected during the backward movement, extract the real-time point cloud cluster corresponding to the locking pin endpoint and calculate its real-time center coordinates.

[0153] In this embodiment, at each moment when the vehicle moves backward slightly, the newly collected point cloud of the locking pin area can be precisely segmented to extract the part of the point cloud cluster representing the locking pin endpoint. It can be understood that this point cloud cluster has been pre-located within the initially extracted ROI. Therefore, only clustering or feature thresholds are needed to quickly screen the locking pin endpoint cluster in the latest data, and calculate the numerical average position of all points within this cluster - that is, the "real-time center coordinates" at this moment. In this way, a series of spatial positions of the locking pin endpoints updated as the vehicle retreats can be obtained.

[0154] Step S18106: Determine whether the position of the locking pin meets the preset stability condition according to the continuous change amount of the real-time center coordinates; wherein, the preset stability condition includes: the displacement change amount of the real-time center coordinates of the locking pin endpoint in the horizontal direction does not exceed the horizontal deviation range; the displacement change amount in the vertical direction does not exceed the vertical deviation range.

[0155] In this embodiment, the real-time central coordinates of the locking pin endpoints calculated in multiple timing frames (for example, 5 - 1 consecutive frames) can be compared pairwise, and the maximum displacements in the horizontal direction (X / Y axes in the track plane) and the vertical direction (Z axis) can be statistically analyzed. If these maximum displacement values do not exceed the preset horizontal / vertical deviation thresholds (for example, horizontal ≤ 1 mm, vertical ≤ 0.5 mm), it is considered that the position of the locking pin endpoint remains extremely stable during the reverse micro-movement process - meaning that the locking pin has been firmly inserted into the pin hole without any loosening or retraction.

[0156] Step S18108: If the real-time central coordinates meet the preset stability condition, it is finally determined that the hooking is successful.

[0157] In this embodiment, only when the locking pin endpoint continuously meets the above-mentioned horizontal and vertical stability conditions during the entire reverse micro-movement verification process, the control device finally confirms that the hooking is completely successful. At this time, the control device will generate a "hooking successful" signal to notify the dispatching and allow the vehicle to release the hooking brake and enter the formation operation; if the stability condition is not met, an abnormal handling process will be triggered, such as an alarm prompt or a re-attempt to hook. Through this final judgment mechanism of multi-frame timing and micro-movement verification, the false locking pin phenomenon caused by vibration, impact, or external force can be effectively excluded, greatly improving the reliability and safety of automatic hooking.

[0158] After initially determining that the "hooking is successful", this final judgment strategy dynamically tracks the position change of the center of the locking pin endpoint by moving the vehicle in reverse and continuously collecting the point cloud of the locking pin area. Only when the displacement fluctuations in the horizontal and vertical directions both remain within the preset small tolerance range, the hooking is finally confirmed to be completed. Such multi-frame timing stability verification can timely capture any risk of locking pin retraction or loosening caused by vibration, inertia, or external force, avoiding false "hooking successful" judgments, thereby significantly improving the reliability and safety of the entire automatic hooking process.

[0159] As Figure 2 shown, this embodiment provides an automatic unhooking method for a rail vehicle, including: Step S20: In response to the received unhooking instruction, send a control signal to the power device of the main coupler of the first rail vehicle to drive the power device to perform a coupler unlocking operation; wherein, the main coupler and the target coupler are in a hooked state; wherein, the first rail vehicle has performed the hooking operation of the main coupler and the target coupler by using the automatic hooking method of a rail vehicle described in the above embodiment, so that the main coupler and the target coupler are in a hooked state.

[0160] Step S22: During the execution of the coupler unlocking operation, three-dimensional point cloud data of the hooking area is collected by a preset lidar, and the real-time point cloud voxel block of the locking pin area is extracted.

[0161] Step S24: Based on the real-time point cloud voxel block and a preset locking pin not-in-place model, a matching comparison is performed to obtain the result of the matching comparison; wherein, the locking pin not-in-place model includes the reference point cloud distribution characteristics after the locking pin is pulled out.

[0162] In this embodiment, the locking pin not-in-place model can be a reference model of the point cloud and geometric features corresponding to the state where the locking pin is jacked up but has not fallen into the pin hole after the unlocking operation is executed. Specifically, the typical vertical offset of the locking pin relative to the hook head reference plane at the release position can be recorded in the model. For example, the absolute height range after rising several centimeters from the depth of the normal landing hole. After the locking pin is pushed out of the pin hole, its rod body and end cap will not be blocked by the hook head hole wall. Therefore, a complete cylindrical surface shape and a clear end face can be presented in the point cloud. The model stores the radius, height, and end face shape of this exposed cylinder in the form of voxels or meshes. Different from the situation where the end point of the locking pin is almost at the center of the hole during normal landing, the locking pin not-in-place model can preset the horizontal and vertical offset vectors of the locking pin upward or outward after unlocking to help the control device quickly locate the position area of the jacked-up locking pin. To support fast matching, the model also includes local geometric descriptors (such as FPFH histograms or SHOT features) of the exposed surface and the cylindrical side surface of the locking pin, which can accurately reflect the normal vector distribution and curvature pattern of the exposed locking pin.

[0163] Step S26: Based on the result of the matching comparison, it is judged whether the locking pin is successfully pulled out.

[0164] Step S28: When it is determined that the locking pin is successfully pulled out, control the first rail vehicle to move backward until the lidar detects that the hooking separation distance is greater than a preset safety distance threshold, and then confirm that the unhooking is completed.

[0165] In this embodiment, while activating the unlocking mechanism, the lidar is used to scan the hooking area in real time and extract the locking pin voxels. By accurately matching with the "not-in-place" model to judge the pin pulling state, it effectively avoids mechanical jamming caused by incomplete unlocking; if it is confirmed that the locking pin has been pulled out, the vehicle can be safely reversed until a sufficient separation distance is scanned to end the process. Throughout the process, no manual intervention is required, which not only ensures the reliability and accuracy of the unhooking action, but also greatly improves the operation safety and system automation level through the early warning and safety distance threshold mechanism.

[0166] In some embodiments, the step of judging whether the locking pin is successfully pulled out based on the result of the matching comparison includes: Step S262: If the voxel blocks of the point cloud in the lock pin area do not cover the predefined lock pin feature area in the lock pin not - in - place model, or the center coordinates of the point cloud cluster corresponding to the lock pin end deviate from the center of the hook hole position by more than the preset unhooking tolerance range, it is determined that the lock pin has been successfully pulled out.

[0167] In this embodiment, the voxel blocks of the lock pin area extracted in real - time can be mapped into the feature area framework defined by the lock pin not - in - place model. This feature area can correspond to the spatial position that the lock pin end may occupy after unlocking. If almost no real - time voxels are detected in this model feature area (i.e., the voxel blocks do not cover this area), it indicates that the lock pin has completely exited this position.

[0168] In this embodiment, it is also possible to identify the point cloud cluster of the lock pin end from the real - time voxels and calculate the geometric center coordinates of this cluster. Then, compare this center with the center of the hook hole position. If the deviation amount in either the horizontal or vertical direction exceeds the preset unhooking tolerance threshold (for example, ±10 mm), it indicates that the lock pin end has moved out of the hole position boundary.

[0169] Step S264: If the center coordinates of the point cloud cluster corresponding to the lock pin end are still within the preset unhooking tolerance range of the hook hole position area, it is determined that the unhooking is not completed, and an unlocking abnormal alarm signal is generated.

[0170] According to an embodiment of the present invention, an electronic device is provided. Please refer to Figure 3 . The electronic device in this embodiment may include one or more of the following components: a processor, a network interface, a memory, a non - volatile memory, and one or more application programs. One or more application programs may be stored in the non - volatile memory and configured to be executed by one or more processors. One or more programs are configured to execute the method described in the foregoing method embodiments.

[0171] According to an embodiment of the present invention, a computer - readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a computer, the computer executes the method described in any of the above - mentioned embodiments.

[0172] According to an embodiment of the present invention, a computer program product containing instructions is also provided. When the instructions are executed by a computer, the computer executes a method in any of the above - mentioned embodiments.

[0173] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of this application are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments of this application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0174] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, and will not be elaborated here.

[0175] The serial numbers of the above embodiments of this application are only for description and do not represent the superiority or inferiority of the embodiments.

[0176] In the above embodiments of this application, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0177] The above are only the preferred embodiments of this application. It should be noted that for those of ordinary skill in the art, without departing from the principle of this application, several improvements and refinements can still be made, and these improvements and refinements should also be regarded as the protection scope of this application.

Claims

1. An automatic coupling method for a rail vehicle, characterized in that, Including: Controlling a first rail vehicle to travel at a preset first speed, and continuously collecting three-dimensional point cloud data in front of the first rail vehicle through a preset lidar during the travel; wherein, the front of the first rail vehicle refers to the front of the body coupler of the first rail vehicle; the lidar is disposed on the first rail vehicle and on the same side as the body coupler; Based on the three-dimensional point cloud data, determining whether there is a target coupler; wherein, the target coupler is configured on one side of a second rail vehicle relative to the body coupler of the first rail vehicle; In the case where it is determined that there is a target coupler, calculating docking deviation data between the target coupler and the body coupler; wherein, the docking deviation data at least includes docking distance deviation data; Based on the distance deviation data and a preset speed control strategy, controlling the traveling speed of the first rail vehicle so that the first rail vehicle approaches the second rail vehicle at a speed not higher than and not equal to the first speed; In the case where it is determined that a preset hooking distance is reached, performing hooking closure.

2. The method according to claim 1, characterized in that, The step of determining whether there is a target coupler based on the three-dimensional point cloud data includes: Performing feature extraction on the three-dimensional point cloud data to obtain feature data for characterizing coupler features; Based on the feature data for characterizing coupler features, determining whether there is a target coupler in front of the body coupler.

3. The method according to claim 1, characterized in that, The docking deviation data further includes: docking angle deviation data; The step of calculating docking deviation data between the target coupler and the body coupler in the case where it is determined that there is a target coupler includes: Based on the latest three-dimensional point cloud data, extracting target three-dimensional point cloud data; wherein, the target three-dimensional point cloud data is three-dimensional point cloud data representing the area where the target coupler is located; Performing a preset surface fitting algorithm on the target three-dimensional point cloud data to generate a fitted quadratic surface model; Based on the quadratic surface model, calculating the center point coordinates and normal vector direction of the target coupler; Based on the pre-calibrated center point coordinates and normal vector direction of the body coupler and the center point coordinates and normal vector direction of the target coupler, calculating distance deviation data and docking angle deviation data.

4. The method according to claim 1, characterized in that The step of controlling the traveling speed of the first rail vehicle based on the distance deviation data and a preset speed control strategy includes: In the case where it is determined that there is a target coupler, switching the first speed to a second speed and controlling the first rail vehicle to travel at a constant speed at the second speed; wherein, the second speed is lower than the first speed; Determining whether the distance deviation data reaches a preset speed switching threshold; In the case where the preset speed switching threshold is reached, switching the second speed to a third speed and controlling the first rail vehicle to travel at the third speed; wherein, the third speed is lower than the second speed.

5. The method according to claim 4, wherein Before controlling the traveling speed of the first rail vehicle based on the distance deviation data and a preset speed control strategy so that the first rail vehicle approaches the second rail vehicle at a speed not higher than and not equal to the first speed, and in the case where the coupler is a Janney coupler, the method further includes: Send a control signal to the power device of the main coupler to open the main coupler so that the main coupler is in a state of waiting for connection; The step of performing hook closing when it is determined that the preset hook-up distance is reached includes: When it is determined that the preset hook-up distance is reached, control the first rail vehicle and the second rail vehicle to touch at a speed not higher than the third speed; After detecting the occurrence of the touch, collect three-dimensional point cloud data of the hook area through a preset lidar; For the three-dimensional point cloud data of the hook area, extract the point cloud voxel block of the locking pin area of the Janney coupler; Based on the extracted point cloud voxel block of the locking pin area and a preset locking pin positioning model, perform a preliminary judgment on whether the hook-up is successful; When the preliminary judgment is that the hook-up is successful, execute a preset final judgment strategy to finally judge whether the hook-up is successful.

6. The method according to claim 5, characterized in that, The step of performing a preliminary judgment on whether the hook-up is successful based on the extracted point cloud voxel block of the locking pin area and a preset locking pin positioning model includes: Match and compare the point cloud voxel block of the locking pin area with the pre-stored reference point cloud spatial distribution in the locking pin positioning model to obtain the spatial alignment relationship between the two; Based on the spatial alignment relationship, extract the point cloud cluster corresponding to the locking pin end point from the point cloud voxel block and calculate the center coordinates of the point cloud cluster; Judge whether the center coordinates of the point cloud cluster fall within the three-dimensional tolerance range of the hook head pin hole area calibrated in the locking pin positioning model, and judge whether the absolute value of the vertical direction coordinate deviation is within the preset vertical deviation range; If both are satisfied, it is judged that the locking pin has been correctly positioned and the preliminary judgment is that the hook-up is successful.

7. The method according to claim 5, characterized in that, The step of performing a preset final judgment strategy to judge whether the hook-up is successful when the preliminary judgment is that the hook-up is successful includes: Control the first rail vehicle to move backward a preset distance along the driving direction, and during the backward movement, continuously collect three-dimensional point cloud data of the locking pin area through the lidar; Based on the three-dimensional point cloud data collected during the backward movement, extract the real-time point cloud cluster corresponding to the locking pin end point and calculate its real-time center coordinates; According to the continuous change amount of the real-time center coordinates, judge whether the position of the locking pin meets the preset stability condition; wherein, the preset stability condition includes: the horizontal displacement change amount of the real-time center coordinates of the locking pin end point does not exceed the horizontal deviation range; the vertical displacement change amount does not exceed the vertical deviation range; If the real-time center coordinates meet the preset stability condition, it is finally determined that the hook-up is successful.

8. An automatic uncoupling method for a rail vehicle, characterized in that, Includes: In response to the received uncoupling instruction, send a control signal to the power device of the main coupler of the first rail vehicle to drive the power device to perform the coupler unlocking operation; wherein, the main coupler and the target coupler are in a hooked state; wherein, the first rail vehicle has performed the hooking operation of the main coupler and the target coupler by using any one of the automatic hooking methods of the rail vehicle described in claims 1-7, so that the main coupler and the target coupler are in a hooked state; During the execution of the coupler unlocking operation, 3D point cloud data of the hook area is collected by a preset lidar, and the real-time point cloud voxel block of the locking pin area is extracted; Based on the real-time point cloud voxel block and a preset locking pin not in place model for matching and comparison, the result of the matching and comparison is obtained; wherein, the locking pin not in place model includes the reference point cloud distribution characteristics after the locking pin is pulled out; Based on the result of the matching and comparison, it is judged whether the locking pin is successfully pulled out; In the case where it is determined that the locking pin is successfully pulled out, the first rail vehicle is controlled to move backward until the lidar detects that the hook separation distance is greater than the preset safety distance threshold, and then the uncoupling is confirmed to be completed.

9. The method according to claim 8, wherein The step of judging whether the locking pin is successfully pulled out based on the result of the matching and comparison includes: If the point cloud voxel block of the locking pin area does not cover the predefined locking pin feature area in the locking pin not in place model, or the center coordinate of the point cloud cluster corresponding to the locking pin end deviates from the center of the hook hole position by more than the preset uncoupling tolerance range, it is determined that the locking pin has been successfully pulled out; If the center coordinate of the point cloud cluster corresponding to the locking pin end is still within the preset uncoupling tolerance range of the hook hole position area, it is determined that the uncoupling is not completed, and an unlocking abnormal alarm signal is generated.

10. An electronic device, characterized in that, Includes: A memory, and one or more processors communicatively connected to the memory; Instructions executable by the one or more processors are stored in the memory, and the instructions are executed by the one or more processors to enable the one or more processors to implement the method according to any one of claims 1 to 7.

Images