A coking rail car anti-collision warning method and system based on multi-source perception

Abstract

This invention belongs to the field of rail transit operation safety and control technology, specifically relating to a collision avoidance warning method and system for coking railcars based on multi-source sensing. The method includes: acquiring the center position coordinates, continuous image sequences, and real-time attitude angles of the coking railcar; identifying visual measurement coordinates and physical coordinates; calculating the spatial position of the visual measurement coordinates through a dynamic transformation matrix for mapping correction to obtain spatial alignment, and acquiring brake loss correction indicators and real-time risk values; determining an elastic alarm distance based on the brake loss correction indicators, real-time risk values, and spatial alignment; and finally determining the safety level based on the relationship between the distance between any coking railcar and any operator and the elastic alarm distance. This invention reduces warning drift caused by installation position deviation and improves the braking safety of heavy-duty railcars under complex operating conditions.

Description

Technical Field

[0001] This invention relates to the field of rail transit operation safety and control technology. More specifically, this invention relates to a collision avoidance early warning method and system for coking railcars based on multi-source sensing. Background Technology

[0002] In the high-risk and complex environment of the coking area of ​​a coking plant, the core challenge of safety management lies in the fact that traditional methods cannot cope with the dynamic and changing risks. The coking railcar includes five major vehicles: coal pusher, coke pusher, coke catcher, coke quencher, and smoke guide. Due to the long hours of operation by the driver, the large size of the vehicles, and the existence of serious blind spots, the driver may not be able to notice the workers who are cleaning, which may lead to the workers being injured by the large vehicles.

[0003] To achieve safe collision avoidance and early warning for rail equipment, ranging sensors and spatial positioning base stations are commonly installed at the ends of vehicles. For example, Chinese patent application CN120503843A discloses a TBM tunnel locomotive positioning and collision avoidance early warning system and method. This system utilizes positioning base stations deployed on the tunnel sidewalls, combined with ranging equipment at the locomotive ends, to acquire locomotive position information and measure relative physical distances. The application also discloses a collision avoidance alarm mechanism triggered based on values ​​collected by the ranging hardware. When the relative distance is less than a fixed safety threshold, the onboard main control system activates an alarm device to remind the driver to brake.

[0004] However, during the full-load start-up and coke pushing operation of coking locomotives, there are problems with rigid warning boundary settings and distorted spatial positioning systems, leading to false alarms by the collision avoidance system. Because track friction dynamically changes with the environment, existing fixed collision avoidance thresholds cannot compensate for distance changes as the car body's kinetic energy decays, resulting in insufficient estimation of the clearance braking buffer margin for heavy-duty locomotives. Simultaneously, the main beam of the car body undergoes non-rigid torsional deformation under heavy loads, and traditional rigid mapping models ignore microscopic geometric rheological physical characteristics. This defect causes the actual coordinates of the ranging antenna to deviate from the initial reference parameters, ultimately resulting in a lag in the physical response of the collision avoidance warning. Summary of the Invention

[0005] To address the aforementioned technical problems of spatial drift and delayed early warning, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a collision avoidance early warning method for coking railcars based on multi-source sensing, comprising: The system acquires continuous image sequences of the center position coordinates and running direction of the coking railcars, as well as their real-time attitude angles. It then performs target recognition on the continuous image sequences to obtain the three-dimensional visual measurement coordinates of each static marker and the three-dimensional physical coordinates of each worker. The three-dimensional visual measurement coordinates are homogeneously expanded to obtain four-dimensional visual measurement coordinates. A dynamic transformation matrix is ​​constructed based on the three-dimensional spatial vector synthesis of the real-time attitude angles. This dynamic transformation matrix is ​​used to map and correct the spatial position of the visual measurement coordinates, resulting in the spatial alignment of each coking railcar. Based on the deviation between the real-time kinetic energy and the standard braking capacity of the coking railcars, the brake loss correction index for each coking railcar is calculated. Combining the theoretical braking distance of any coking railcar, the distance to any worker, and the spatial alignment, the real-time risk value of the corresponding coking railcar for any worker is obtained. Based on the brake loss correction index, the real-time risk value, the spatial alignment, and the spatial deviation of the real-time attitude angles, the elastic alarm distance between the corresponding coking railcar and the corresponding worker is determined. Finally, based on the relationship between the distance between any coking railcar and any worker and the elastic alarm distance, the safety level of the current working condition is determined.

[0007] This invention comprehensively acquires the center position coordinates of the coking railcar, continuous image sequences, and real-time attitude angles. It uses the three-dimensional spatial vector synthesis of the real-time attitude angles to construct a dynamic transformation matrix and calculates the spatial position mapping correction of the visual measurement coordinates of static markers using this matrix. This transforms the geometric deviation caused by the non-rigid torsion of the car body into an assessable spatial alignment. Combined with the deviation between the real-time kinetic energy of the coking railcar and the standard braking capacity, a brake loss correction index is obtained. This, along with the real-time risk value, determines the elastic alarm distance, enabling the warning threshold to be dynamically adjusted according to the decay of mechanical performance and changes in the perceived environment. This enhances the adaptive capability of the warning system in complex coking operation environments such as heavy loads, vibrations, and coal dust obstruction, thereby reducing the judgment lag caused by fixed alarm distances and improving the safety redundancy of mobile equipment operation within the coking plant.

[0008] Preferably, the acquisition of the center position coordinates, continuous image sequence of the running direction, and real-time attitude angle of the coking railcar includes: The center position coordinates of each coking railcar are obtained using a positioning terminal; a continuous image sequence of the running direction of each coking railcar is acquired using an RGB-D camera; a deformation monitoring network is formed by arranging nine-axis inertial measurement units at the main beam support, motor torque support seat, and bogie crossbeam of each coking railcar; the angular velocity and acceleration signals at each structural node in the deformation monitoring network are continuously read; and the integral result of the angular velocity is fused with the gravity vector angle calculated from the acceleration signal to obtain the real-time attitude angle.

[0009] This invention forms a deformation monitoring network by arranging nine-axis inertial measurement units at structural nodes such as the main beam support, motor torque support seat, and bogie crossbeam. The real-time attitude angle is obtained by fusing the angular velocity integral result with the gravity vector angle calculated from the acceleration signal. This allows for real-time monitoring of the structural displacement changes of the coking railcar when it is carrying coke pushing or coal loading loads. This improves the initial accuracy of multi-source sensing data under severe mechanical impact conditions, enhances the reliability of subsequent spatial coordinate alignment, reduces the distortion of the sensing coordinate system caused by the flexible deformation of the car body structure, and improves the system's sensitivity to deformation monitoring of coking equipment.

[0010] Preferably, the spatial alignment of the coking railcar satisfies the following relationship: ; In the formula, Indicates the first A coking railcar Spatial alignment at any given moment; Indicates static signage in Visual coordinates of the moment; Indicates the first A coking railcar The dynamic transformation matrix at time step; Indicates static signage in The known standard coordinates of the four dimensions at time; This indicates the preset sensitivity adjustment coefficient; The symbol for the Euclidean norm of a vector.

[0011] This invention utilizes the visual measurement coordinates of static markers and the spatial alignment degree calculated by the reprojection error of known standard coordinates in the three-dimensional space under the geographic coordinate system. It presents the confidence level of the visual perception system after being affected by environmental interference and vehicle deformation in a numerical form, reflecting the true matching state between the perception result and the coordinates of the physical world. This improves the early warning system's ability to identify abnormal fluctuations in sensor data, reduces false alarms caused by data drift, and enhances the robustness of the perception system in the complex dust environment of a coking plant.

[0012] Preferably, the brake loss correction index of the coking railcar satisfies the following relationship: ; In the formula, Indicates the first A coking railcar Braking wear correction index at all times; Indicates the first A coking railcar The quality of time; Indicates the first A coking railcar The speed of operation at any given moment; Indicates the coefficient of kinetic friction; Indicates the first Standard load mass of a coking railcar; Represents gravitational acceleration; Indicates the standard braking distance.

[0013] This invention compares the real-time mechanical energy of a coking railcar with the ultimate work done by the rail friction within the standard braking displacement to obtain a brake loss correction index. It incorporates load mass, running speed, and rail lubrication status characterized by the dynamic friction coefficient into the braking performance evaluation model, reflecting the energy dissipation pressure of the braking system under different operating conditions. This improves the compatibility of the safety boundary with different material loading plans and weather conditions, reduces the phenomenon of insufficient braking distance prediction caused by ignoring the limitations of the actual operating environment and dynamic inertia changes, and ensures that the railcar still has sufficient braking buffer margin in rainy, snowy, or oily rail conditions with friction attenuation.

[0014] Preferably, the real-time risk value of the coking railcar for any operator satisfies the following relationship: ; In the formula, Indicates the first Coking railcars for the first One worker Real-time risk value at any given moment; Indicates the first A coking railcar The speed of operation at any given moment; Indicates the coefficient of kinetic friction; Represents gravitational acceleration; Indicates the first Coking railcars for the first One worker The distance of time; Indicates the first A coking railcar Spatial alignment at any given moment.

[0015] Preferably, obtaining the elastic alarm distance includes: The ratio of the sum of the brake loss correction index of any coking railcar at the current moment, the real-time risk value of any operator, and 1, to the product of the spatial alignment of the corresponding coking railcar at the current moment and the cosine of the equivalent deflection angle, is multiplied by the standard braking distance to obtain the elastic alarm distance of the corresponding coking railcar to the corresponding operator at the current moment.

[0016] This invention obtains the elastic alarm distance of the coking railcar to the operators at the current moment, and makes the warning boundary expand and contract in real time with the increase of physical collision risk and the loss of sensing geometric projection. When the perception confidence decreases or the equivalent deflection angle of the vehicle increases, the elastic alarm distance is forcibly amplified by reducing the denominator value, thereby achieving distance compensation for the loss of field of view and coordinate offset of the sensing system. This improves the response sensitivity of the collision avoidance warning in critical dangerous conditions, balances the contradiction between production efficiency and emergency braking intensity, reduces the collision risk caused by the narrow fixed threshold, and enhances the safety defense depth of the coking railcar in dynamic and complex scenarios.

[0017] Preferably, determining the safety level of the current working condition based on the relationship between the distance between any coking railcar and any operator and the elastic alarm distance includes: The distance between any coking railcar and any worker at the current moment is compared with the elastic alarm distance between the corresponding coking railcar and the corresponding worker at the current moment. When the distance is less than or equal to the elastic alarm distance, it is determined to be a high-risk collision condition. The highest priority emergency braking command is immediately sent to the hydraulic braking actuator through the multi-channel PID controller to force the corresponding coking railcar to stop in the shortest possible time, and the audible and visual alarm device is activated simultaneously to remind the corresponding worker to avoid the collision. When the distance is greater than the elastic alarm distance, it is determined to be a safe condition, and the current operation command remains unchanged.

[0018] Preferably, obtaining the dynamic transformation matrix includes: The real-time attitude angle of any coking railcar at the current moment is compared with the pre-stored static reference angle to obtain the attitude angle deviation of the corresponding coking railcar in the roll, pitch and yaw directions at the current moment. Calculate the Euclidean norm of the attitude angle deviations in the three directions of roll, pitch and yaw to obtain the equivalent deflection angle of the coking railcar at the current moment. A corrected rotation matrix is ​​constructed based on the equivalent deflection angle. The corrected rotation matrix is ​​then multiplied by the initially calibrated reference matrix to obtain the dynamic transformation matrix of the corresponding coking railcar at the current moment.

[0019] Preferably, the step of performing target recognition on the continuous image sequence to obtain the three-dimensional visual measurement coordinates of each static marker and the three-dimensional physical coordinates of each worker includes: Human features are extracted from continuous image sequences of each coking railcar's running direction based on pre-trained YOLO series object detection algorithms. Combined with SIFT local feature matching algorithm, static markers along the track are geometrically located to obtain the two-dimensional coordinates of each static marker and each worker. Then, the true physical depth value in the depth channel of the RGB-D camera is read synchronously and mapped and transformed using the intrinsic parameter matrix of the RGB-D camera to obtain the three-dimensional visual measurement coordinates of each static marker and the three-dimensional physical coordinates of each worker.

[0020] Secondly, the present invention provides a coking railcar collision avoidance and early warning system based on multi-source perception, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned coking railcar collision avoidance and early warning method based on multi-source perception is implemented.

[0021] By adopting the above technical solution, a computer program for a coking railcar collision avoidance early warning method based on multi-source sensing is generated and stored in a memory for loading and execution by a processor. Terminal devices are then created based on the memory and processor for convenient use.

[0022] The beneficial effects of this invention are as follows: Addressing the spatiotemporal overlap and conflict caused by the randomness of personnel movement and the operation of rail vehicles in a collaborative coking production environment, this invention integrates multi-physics sensor data to obtain vehicle posture, uses spatial position mapping to correct visual deviations caused by non-rigid torsion and dust obstruction, and obtains spatial alignment. It combines the deviation of the vehicle's real-time kinetic energy and standard braking performance to obtain brake loss correction indicators, assesses the real-time risk value of human-vehicle interaction based on the degree of mechanical braking limitation, and obtains an elastic alarm distance that expands and contracts in real-time with environmental conditions and vehicle performance through multi-dimensional feature integration. This enables the collision avoidance system to adaptively identify the attenuation of frictional force on oil-stained tracks and the inertial impact caused by heavy loads. When the perceived quality deteriorates, it automatically amplifies the warning depth to compensate for spatial losses. Thus, while maintaining the continuous operational efficiency of the coking plant, it reduces production interference caused by imbalanced warning threshold settings, lowers the probability of physical collisions caused by insufficient buffer space estimation in emergency situations, and enhances the dynamic safety defense depth of mixed operation of mobile equipment and on-site personnel. Attached Figure Description

[0023] Figure 1 This is a flowchart of a collision avoidance and early warning method for coking railcars based on multi-source sensing; Figure 2 This is a schematic diagram of the initial spatial distribution; Figure 3 This is a diagram illustrating the real-time risk value distribution; Figure 4 This is a diagram illustrating the elastic alarm distance. Detailed Implementation

[0024] This invention discloses a collision avoidance and early warning method for coking railcars based on multi-source sensing, referring to... Figure 1 This includes steps S100-S500: S100, multi-source sensing data acquisition and 3D physical coordinate extraction.

[0025] It should be noted that since a single-dimensional sensor cannot maintain stable sensing accuracy under harsh working conditions, it is necessary to use time-dimensional filtering and smoothing and spatial multi-source signal complementarity to cancel environmental noise. Therefore, this invention uses a positioning terminal, an RGB-D camera and a nine-axis inertial measurement unit to collect signals, and combines Kalman filtering and complementary filtering algorithms to purify the original signals, thereby suppressing the interference of random noise on the initial pose data.

[0026] Specifically, the center position coordinates of each coking railcar are obtained using a positioning terminal, and these coordinates are then smoothed. For example, the smoothing process employs Kalman filtering, which can reduce positioning data jumps caused by multipath effects. It should be noted that coking railcars typically include five types: coal pusher, coke pusher, coke quencher, coke extinguishing car, and flue gas guide car.

[0027] A continuous image sequence of each coking railcar's running direction is acquired using an RGB-D camera. This sequence is then processed for image enhancement, noise reduction, and target recognition to obtain the two-dimensional coordinates of each static marker and each worker. Simultaneously, the actual physical depth value from the RGB-D camera's depth channel is read and mapped using the camera's intrinsic parameter matrix to obtain the three-dimensional visual measurement coordinates of each static marker and the three-dimensional physical coordinates of each worker. For example, the image enhancement uses histogram equalization, and the noise reduction uses nonlocal mean filtering. Histogram equalization enhances the overall image contrast and highlights target details in strong or weak light conditions; while nonlocal mean filtering reduces the complex noise of the coking plant while more completely preserving edge features.

[0028] It should be noted that the target recognition is based on extracting human features from continuous image sequences using a pre-trained deep neural network model, and combining this with a feature matching algorithm to geometrically locate static markers along the track, thereby obtaining the visual measurement coordinates of each static marker and the physical coordinates of each worker. For example, the deep neural network model is the YOLO series target detection algorithm, and the feature matching algorithm is the SIFT local feature matching algorithm. Both the YOLO series target detection algorithm and the SIFT local feature matching algorithm are existing technologies in the field of computer vision and will not be elaborated upon here.

[0029] A deformation monitoring network is formed by arranging nine-axis inertial measurement units at structural nodes such as the main beam supports, motor torque support seats, and bogie crossbeams of each coking railcar. Angular velocity and acceleration signals at each structural node in the deformation monitoring network are continuously read. The integral result of the angular velocity is fused with the gravity vector angle calculated from the acceleration signal to obtain the real-time attitude angle. It should be noted that the complementary filtering algorithm is existing technology and will not be elaborated upon here.

[0030] For example, Figure 2 This is an initial spatial distribution diagram, which shows the physical positional relationship between each coking railcar and each operator at a specific operating moment. By presenting the spatial distribution between multiple coking railcars and dispersed operators, it reflects the randomness and potential spatiotemporal overlap and conflict in the collaborative operation environment of the coking plant.

[0031] This completes the acquisition and preprocessing of multi-source sensing data.

[0032] S200, Calculation of coordinate extension and spatial alignment in 3D vision measurement.

[0033] It should be noted that during the process of fully loading coal or pushing coke, the main beam of the vehicle body will undergo torsional deformation, causing a dynamic deviation between the camera mounting coordinate system fixed on the vehicle body and the track geographic coordinate system. Considering that the preset static markers have fixed physical spatial attributes, their projection drift in the image can directly characterize the degree of vehicle body deformation. Therefore, this invention establishes a coordinate alignment model based on a dynamic transformation matrix and uses reprojection error as a variable to construct an evaluation function, thereby alleviating the problem of perceptual spatial distortion caused by vehicle body structural deformation.

[0034] Specifically, based on the spatial registration residual between the corrected visual predicted pose and the global standard pose, the spatial alignment of each coking railcar is calculated, including: The real-time attitude angle of any coking railcar at the current moment is compared with the pre-stored static reference angle to obtain the attitude angle deviation of the coking railcar in the roll, pitch, and yaw directions at the current moment. For example, the pre-stored static reference angle is the initial attitude angle data continuously collected by the deformation monitoring network when the coking railcar is in an unloaded and stationary state and is located in a standard horizontal track section. It is used to characterize the standard spatial attitude of the coking railcar without mechanical torsional deformation.

[0035] Calculate the Euclidean norm of the attitude angle deviations in the roll, pitch, and yaw directions to obtain the equivalent deflection angle of the corresponding coking railcar at the current moment.

[0036] A corrected rotation matrix is ​​constructed based on the equivalent deflection angle. Multiplying this corrected rotation matrix by the initially calibrated reference matrix yields the dynamic transformation matrix of the corresponding coking railcar at the current moment. It should be noted that this dynamic transformation matrix is ​​a 4×4 homogeneous transformation matrix.

[0037] After obtaining the position values ​​of the three directions of the three-dimensional visual measurement coordinates of each static marker at the current moment, a constant with a value of one is added as a fourth auxiliary dimension, thereby expanding the original three-dimensional coordinates vertically into a four-dimensional homogeneous column vector form, which serves as the visual measurement coordinates of the corresponding static marker at the current moment.

[0038] It should be further explained that the dynamic transformation matrix is ​​used to map the visual measurement coordinates to a unified geographic coordinate system to compensate for the angular distortion caused by the non-rigid torsion of the coking railcar during operation, and to obtain the predicted coordinates of the target transformed to the unified geographic coordinate system. The three-dimensional spatial reprojection error between the predicted coordinates and the known four-dimensional standard coordinates is further calculated. The smaller the three-dimensional spatial reprojection error, the more complete the compensation for the deformation of the vehicle body, the higher the alignment between visual perception and the physical world, and the higher the spatial alignment obtained in the end. This characterizes the dynamic impact of extreme working conditions on the collision avoidance positioning accuracy of the system.

[0039] The spatial alignment of any coking railcar satisfies the following relationship: ; In the formula, Indicates the first A coking railcar Spatial alignment at any given moment; Indicates static signage in Visual coordinates of the moment; Indicates the first A coking railcar The dynamic transformation matrix at time step; Indicates static signage in The known standard coordinates of the four dimensions at any given time are statically calibrated based on the preset physical location of the static marker in the geographical coordinate system of the coking plant, and the values ​​are retrieved from the global map database pre-stored in the system. This represents the preset sensitivity adjustment coefficient, set to 10m, which is a spatial deviation scaling value determined based on the measurement accuracy of the RGB-D camera and the tolerance of environmental noise. The symbol for the Euclidean norm of a vector.

[0040] In this relation, This indicates that the visual measurement coordinates are transformed into predicted coordinates under the unified geographic coordinate system of the coking plant through a dynamic transformation matrix. This represents the three-dimensional spatial reprojection error between the visually predicted coordinates after deformation compensation and the actual standard coordinates of the sign. The larger the value, the more likely it is to be the first. Incomplete deformation compensation of the coking railcar body, or severe environmental interference to the sensors, leads to a misalignment between the perceived results and the actual physical space, resulting in lower spatial alignment; conversely, it indicates that the first... The dynamic transformation matrix of a coking railcar can reduce the deviation caused by the torsion of the car body, and the visual perception is aligned with the physical world, with a higher degree of spatial alignment.

[0041] Thus, the spatial alignment of each coking railcar was obtained.

[0042] S300, calculate the brake wear correction index.

[0043] It should be noted that, due to the large mass and operational inertia of the coking railcar, its braking process is not instantaneous. Furthermore, oil stains and water accumulation on the track surface can cause the actual braking distance to far exceed the theoretical design value. Considering that the current kinetic energy of the vehicle and the upper limit of the negative work that the track friction can provide together determine the physical safety limit, this invention introduces a graded adjustment mechanism for the dynamic friction coefficient and uses the ratio of actual kinetic energy to standard work to construct a braking loss correction model, converting the kinetic energy surplus into a distance compensation coefficient, thereby improving the safety redundancy of the braking system under different operating conditions.

[0044] Specifically, based on the energy efficiency matching degree between real-time operating kinetic energy and standard braking capacity, the brake loss correction index for each coking railcar is calculated, including: Set the dynamic friction coefficient. For example, in rainy or snowy weather, when there is a lot of oil on the track surface, or when the ambient humidity is extremely high, the track lubrication is high, the frictional resistance is significantly reduced, and the braking performance is reduced. The dynamic friction coefficient is set to 0.2. Under normal operating conditions where the weather is dry, the track surface is clean, and there is no debris mixed in, the frictional torque between the wheel and the track is sufficient, and the braking effect meets expectations. The dynamic friction coefficient is set to 0.35.

[0045] The first derivative of time is calculated for the smoothed center position coordinates to obtain the real-time running speed of each coking railcar.

[0046] It should be further explained that this invention constructs a brake loss correction index relationship based on the classical kinetic energy theorem. The numerator is calculated by multiplying the mass by the square of the running speed, and then multiplying by 0.5 to obtain the real-time mechanical energy. The denominator is calculated by multiplying the dynamic friction coefficient, standard load mass, gravitational acceleration, and standard braking distance to obtain the system's limit work capacity. By calculating the ratio of real-time mechanical energy to limit work capacity, the brake loss correction index is obtained, thereby measuring the degree of overdraft of rated braking capacity under the current operating conditions and characterizing the additional requirement for collision avoidance safety distance caused by dynamic inertia due to heavy load or speed changes.

[0047] The brake loss correction index of any coking railcar satisfies the following relationship: ; In the formula, Indicates the first A coking railcar Braking wear correction index at all times; Indicates the first A coking railcar The quality at any given moment is calculated by adding the weight of the coking railcar itself to the weight of the material in the current coal loading or coking plan. Indicates the first A coking railcar The speed of operation at any given moment; Indicates the coefficient of kinetic friction; Indicates the first The standard load mass of a coking railcar is determined by the design specifications of the coking railcar; This represents the acceleration due to gravity, usually taken as 9.8 m / s². This represents the standard braking distance, which is the reference physical length required for a vehicle to come to a complete stop under standard load and speed, preset according to the coking operation procedures.

[0048] In this relation, Indicates the first The real-time mechanical energy stored by a coking railcar in the current instantaneous state. The larger the value, the heavier the vehicle's load or the faster its operating speed. This means that the braking system needs to consume and convert a larger amount of energy. With limited friction, this means that the vehicle's braking inertia is extremely large, and the physical cost required to stop is higher. Conversely, it means that the vehicle is under light load or at low speed. The braking system only needs to consume less energy to stop the vehicle, resulting in a high degree of safety redundancy. Indicates the first The limit of work that a coking railcar can perform under ideal design conditions, within the standard braking displacement, is achieved by the rail friction. The larger the value, the higher the hardware performance or design redundancy, or the better the environmental friction conditions; conversely, it indicates limited braking ability or extremely harsh environment.

[0049] Thus, the brake wear correction index for each coking railcar was obtained.

[0050] S400 calculates real-time risk values ​​and elastic alarm distances.

[0051] It should be noted that in a collaborative work environment, the random movement trajectory of the workers and the travel path of the railcar have uncertain spatiotemporal overlap, and the manual avoidance behavior is unpredictable. Considering that the collision risk depends not only on the absolute physical span between the workers and the vehicles, but also on the confidence level of the sensors in the current posture, this invention combines theoretical stopping distance and perception error gain to improve the accuracy of identifying potential dangerous states.

[0052] Specifically, based on the combined risk level of vehicle braking inertia and perception deviation, the real-time risk value of each coking railcar to each operator is calculated, including: The first The physical coordinates of each worker are mapped to the global geographic coordinate system through a dynamic transformation matrix, and then compared with the coordinates of the first worker. Calculate the Euclidean distance from the center coordinates of the coking railcars to obtain the first... Coking railcars for the first The distance between each worker.

[0053] It should be further explained that the real-time risk value is calculated by combining the kinematic braking theory with the uncertainty penalty mechanism in the intelligent control system. The ratio of the theoretical limit braking displacement to the actual distance is obtained by dividing the running speed by the product of the dynamic friction coefficient, the gravitational acceleration, and twice the distance between the person and the vehicle. Since perception inaccuracy will worsen the physical risk with a multiplier effect, the above ratio is further multiplied by the reciprocal of the spatial alignment. The real-time risk value is obtained through nonlinear safety gain compensation. When the required braking space approaches the actual distance or the spatial alignment decreases, the real-time risk value is amplified, thereby characterizing the urgency of obstacle avoidance under extreme conditions.

[0054] The real-time risk value of any coking railcar to any operator satisfies the following relationship: ; In the formula, Indicates the first Coking railcars for the first One worker Real-time risk value at any given moment; Indicates the first A coking railcar The speed of operation at any given moment; Indicates the coefficient of kinetic friction; Represents gravitational acceleration; Indicates the first Coking railcars for the first One worker The distance of time; Indicates the first A coking railcar Spatial alignment at any given moment.

[0055] In this relation, This represents the ratio of the theoretical braking distance due to limited braking capacity to the actual physical distance between the pedestrian and the vehicle. The larger the value, the faster the vehicle is moving or the closer it is to the workers, causing the theoretical stopping distance to approach or even exceed the actual distance, posing a significant risk to mechanical braking. Conversely, the smaller the value, the slower the vehicle is moving or the distance between the vehicle and the workers is sufficient, meaning that even in the event of emergency braking, the vehicle has enough buffer space to stop in front of the workers, indicating a high degree of physical safety redundancy. Indicates that for the first Risk gain coefficient of spatial alignment error in the sensing system of a coking railcar. The larger the value, the more likely the vehicle body twisting has caused a coordinate alignment shift, meaning the system cannot trust the current perception results and must force the system to enter an early warning state by increasing the risk assessment. Conversely, the smaller the value, the more likely the data is true and reliable, and normal judgments can be made based on physical distance without the need for additional risk gain compensation.

[0056] For example, Figure 3 This is a schematic diagram of the real-time risk value distribution. The diagram shows the degree of danger of human-vehicle interaction for each group, as assessed by the system based on braking inertia and perception effectiveness. The overall distribution shows local differences. Most human-vehicle combinations are in a safe state, while the real-time risk value increases between specific coking railcars and specific workers due to the shortened physical distance or decreased spatial alignment.

[0057] At this point, the real-time risk values ​​for each coking railcar and each operator were obtained.

[0058] It should be noted that if the warning threshold is set to a fixed extreme value, it will be difficult for the system to balance between sensitivity and false alarm rate, resulting in frequent false alarm interference or response lag in emergency conditions. This invention combines the real-time kinematic parameters of the coking railcar, braking performance loss and the confidence of the sensing system to reflect the vehicle's real obstacle avoidance needs in a specific posture, thereby improving the adaptive capability of the warning boundary to complex working environments.

[0059] It should be further explained that, using the standard braking distance as the physical benchmark, the numerator is calculated by adding the constant, the brake wear correction index, and the real-time risk value to form a dynamic compensation term that characterizes mechanical overdraft and imminent danger; the denominator is calculated by multiplying the spatial alignment by the absolute value of the cosine of the equivalent deflection angle to characterize the effective perception confidence under non-rigid torsion of the vehicle body. By dividing the dynamic compensation term by the effective perception confidence, when braking performance declines, physical risk increases sharply, or perspective distortion causes a decrease in perception confidence, the calculation result will be forcibly stretched to the alarm threshold, thereby compensating for the extreme braking buffer depth under extremely complex working conditions.

[0060] Preferably, based on vehicle performance loss, personnel collision risk, and error compensation caused by image deflection, the elastic alarm distance of each coking railcar to each operator is calculated. The elastic alarm distance of any coking railcar to any operator satisfies the following relationship: ; In the formula, Indicates the first Coking railcars for the first One worker Flexible alarm distance at any time; Indicates the standard braking distance; Indicates the first A coking railcar Braking wear correction index at all times; Indicates the first Coking railcars for the first One worker Real-time risk value at any given moment; Indicates the first A coking railcar Spatial alignment at any given moment; Indicates the first A coking railcar The equivalent deflection angle at any given moment.

[0061] In this relation, Indicates the first The increase in comprehensive dynamic risks faced by each coking railcar The larger the value, the heavier the braking load on the vehicle and the higher the urgency of the collision. The system needs a larger braking buffer zone, which will lead to a longer elastic alarm distance. Conversely, it indicates that the vehicle is running smoothly, the braking redundancy is sufficient, and the distance to personnel is still far. The system maintains a lower elastic alarm distance, which is conducive to ensuring production efficiency. Indicates the first Effective sensing geometric projection accuracy of a coking railcar. The larger the value, the more reliable the data and the more stable the vehicle's posture. In this case, the original calculation result can be trusted, and no additional safety premium compensation is needed. Conversely, it indicates that the vehicle body is severely tortuous or that the sensing data has a serious drift, resulting in a smaller denominator and thus increasing the elastic alarm distance.

[0062] For example, Figure 4 This is a schematic diagram of the elastic alarm distance. The diagram shows the expansion and contraction of the safety collision avoidance boundary dynamically defined for each group of people and vehicles after the system integrates multi-dimensional features. It not only reflects the targeted extension of the warning depth as the local collision risk increases, but also shows that the collision avoidance baseline for some coking railcars is comprehensively extended due to the overall high brake wear correction index or deformation deviation of the sensing.

[0063] Thus, the flexible alarm distance for each coking railcar to each operator was obtained.

[0064] S500 determines the safety level of the current operating condition by combining the elastic alarm distance.

[0065] It should be noted that the warning threshold fluctuates depending on vehicle status and environmental changes. When the actual physical distance between the personnel and the vehicle decreases to the aforementioned elastic alarm distance, it means that the remaining buffer distance is insufficient to support normal deceleration operations. At this point, extreme actions must be taken to counteract the inertial impulse and prevent a collision. Therefore, by comparing the distance between the coking railcar and the workers with the aforementioned elastic alarm distance, and by driving a multi-channel PID controller to intervene in the hydraulic braking system, the time delay from risk perception to mechanical activation can be shortened.

[0066] Specifically, comparing the impact of any coking railcar on any operator... The distance at a given moment and the corresponding coking railcar for the corresponding workers The system determines the time-sensitive, flexible alarm distance. When the distance is less than or equal to the flexible alarm distance, it is considered a high-risk collision condition. The system immediately sends a high-priority emergency braking command to the hydraulic brake actuator via the multi-channel PID controller, forcing the corresponding coking railcar to stop in the shortest possible time. Simultaneously, the system activates the audible and visual alarm device to alert the corresponding personnel to avoid the collision. When the distance is greater than the flexible alarm distance, it is considered a safe condition, and the current operating command remains unchanged.

[0067] This completes the collision warning system for the coking railcar.

[0068] This invention also discloses a multi-source sensing-based coking railcar collision avoidance and early warning system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a multi-source sensing-based coking railcar collision avoidance and early warning method according to this invention.

[0069] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0070] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A collision avoidance and early warning method for coking railcars based on multi-source sensing, characterized in that, include: Acquire continuous image sequences of the center position coordinates and running direction of the coking railcar, as well as its real-time attitude angle; Target recognition is performed on the continuous image sequence to obtain the three-dimensional visual measurement coordinates of each static marker and the three-dimensional physical coordinates of each worker; The three-dimensional visual measurement coordinates are homogeneously expanded to obtain four-dimensional visual measurement coordinates; a dynamic transformation matrix is ​​constructed based on the three-dimensional spatial vector synthesis of the real-time attitude angles; the spatial position of the visual measurement coordinates is mapped and corrected through the dynamic transformation matrix to obtain the spatial alignment of each coking railcar. Based on the degree of deviation between the real-time kinetic energy and the standard braking capacity of the coking railcar, the brake loss correction index of each coking railcar is calculated. By combining the theoretical braking distance of any coking railcar, the distance to any worker, and the spatial alignment, the real-time risk value of the corresponding coking railcar to any worker is obtained. Based on the brake loss correction index, the real-time risk value, the spatial alignment degree, and the spatial deviation of the real-time attitude angle, the elastic alarm distance between the corresponding coking railcar and the corresponding operator is determined. The safety level of the current working condition is determined based on the relationship between the distance between any coking railcar and any worker and the elastic alarm distance.

2. The method for collision avoidance and early warning of coking railcars based on multi-source sensing according to claim 1, characterized in that, The acquisition of the center position coordinates, continuous image sequence of the running direction, and real-time attitude angle of the coking railcar includes: The center position coordinates of each coking railcar are obtained using a positioning terminal; a continuous image sequence of the running direction of each coking railcar is acquired using an RGB-D camera; a deformation monitoring network is formed by arranging nine-axis inertial measurement units at the main beam support, motor torque support seat, and bogie crossbeam of each coking railcar; the angular velocity and acceleration signals at each structural node in the deformation monitoring network are continuously read; and the integral result of the angular velocity is fused with the gravity vector angle calculated from the acceleration signal to obtain the real-time attitude angle.

3. The method for collision avoidance and early warning of coking railcars based on multi-source sensing according to claim 1, characterized in that, The spatial alignment of the coking railcar satisfies the following relationship: ； In the formula, Indicates the first A coking railcar Spatial alignment at any given moment; Indicates static signage in Visual coordinates of the moment; Indicates the first A coking railcar The dynamic transformation matrix at time step; Indicates static signage in The known standard coordinates of the four dimensions at time; This indicates the preset sensitivity adjustment coefficient; The symbol for the Euclidean norm of a vector.

4. The method for collision avoidance and early warning of coking railcars based on multi-source sensing according to claim 1, characterized in that, The brake loss correction index of the coking railcar satisfies the following relationship: ； In the formula, Indicates the first A coking railcar Braking wear correction index at all times; Indicates the first A coking railcar The quality of time; Indicates the first A coking railcar The speed of operation at any given moment; Indicates the coefficient of kinetic friction; Indicates the first Standard load mass of a coking railcar; Represents gravitational acceleration; Indicates the standard braking distance.

5. A method for collision avoidance early warning of coking railcars based on multi-source sensing according to claim 1, characterized in that, The real-time risk value of the coking railcar for any operator satisfies the following relationship: ； In the formula, Indicates the first Coking railcars for the first One worker Real-time risk value at any given moment; Indicates the first A coking railcar The speed of operation at any given moment; Indicates the coefficient of kinetic friction; Represents gravitational acceleration; Indicates the first Coking railcars for the first One worker The distance of time; Indicates the first A coking railcar Spatial alignment at any given moment.

6. The method for collision avoidance and early warning of coking railcars based on multi-source sensing according to claim 1, characterized in that, The acquisition of the elastic alarm distance includes: The ratio of the sum of the brake loss correction index of any coking railcar at the current moment, the real-time risk value of any operator, and 1, to the product of the spatial alignment of the corresponding coking railcar at the current moment and the cosine of the equivalent deflection angle, is multiplied by the standard braking distance to obtain the elastic alarm distance of the corresponding coking railcar to the corresponding operator at the current moment.

7. A method for collision avoidance and early warning of coking railcars based on multi-source sensing according to claim 1, characterized in that, The determination of the safety level of the current working condition based on the relationship between the distance between any coking railcar and any worker and the elastic alarm distance includes: The distance between any coking railcar and any worker at the current moment is compared with the elastic alarm distance between the corresponding coking railcar and the corresponding worker at the current moment. When the distance is less than or equal to the elastic alarm distance, it is determined to be a high-risk collision condition. An emergency braking command is sent to the hydraulic brake actuator through the multi-channel PID controller to force the corresponding coking railcar to stop in the shortest possible time, and the audible and visual alarm device is activated simultaneously to remind the corresponding worker to avoid the collision. When the distance is greater than the elastic alarm distance, it is determined to be a safe condition, and the current operation command remains unchanged.

8. A method for collision avoidance and early warning of coking railcars based on multi-source sensing according to claim 1, characterized in that, The acquisition of the dynamic transformation matrix includes: The real-time attitude angle of any coking railcar at the current moment is compared with the pre-stored static reference angle to obtain the attitude angle deviation of the corresponding coking railcar in the roll, pitch and yaw directions at the current moment. Calculate the Euclidean norm of the attitude angle deviations in the three directions of roll, pitch and yaw to obtain the equivalent deflection angle of the coking railcar at the current moment. A corrected rotation matrix is ​​constructed based on the equivalent deflection angle. The corrected rotation matrix is ​​then multiplied by the initially calibrated reference matrix to obtain the dynamic transformation matrix of the corresponding coking railcar at the current moment.

9. A method for collision avoidance and early warning of coking railcars based on multi-source sensing according to claim 1, characterized in that, The step of performing target recognition on the continuous image sequence to obtain the three-dimensional visual measurement coordinates of each static marker and the three-dimensional physical coordinates of each worker includes: Human features are extracted from continuous image sequences of each coking railcar's running direction based on pre-trained YOLO series object detection algorithms. Combined with SIFT local feature matching algorithm, static markers along the track are geometrically located to obtain the two-dimensional coordinates of each static marker and each worker. Then, the true physical depth value in the depth channel of the RGB-D camera is read synchronously and mapped and transformed using the intrinsic parameter matrix of the RGB-D camera to obtain the three-dimensional visual measurement coordinates of each static marker and the three-dimensional physical coordinates of each worker.

10. A coking railcar collision avoidance and early warning system based on multi-source sensing, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a coking railcar collision avoidance early warning method based on multi-source sensing according to any one of claims 1-9.

Images