Anti-collision early warning system for remote control operation of engineering machinery

Abstract

The invention discloses an anti-collision early warning system for remote control operation of engineering machinery, and relates to the technical field of anti-collision early warning. A laser radar, a millimeter wave radar and a binocular camera are deployed to form a 360-degree sensing network, and data are synchronized through a GPS timestamp; an improved YOLOv5 algorithm and a density adaptive DBSCAN algorithm are adopted to identify engineering machinery, personnel and obstacles and extract parameters; calculating a collision risk index based on trajectory prediction, and dividing mild, moderate and high risk levels; the early warning decision module adopts a fuzzy control algorithm to generate three-level early warning from visual prompt to automatic deceleration; a 5G and WiFi (Wireless Fidelity) double-link redundancy framework is constructed, and data transmission is guaranteed; developing a console and a tactile feedback handle, presenting risks and providing tactile prompts. According to the method, blind areas are eliminated through source sensing fusion, and targets are accurately recognized; false alarm and missing alarm are reduced through dynamic risk assessment; the safety is improved through three-level early warning and automatic intervention; operation experience is optimized through tactile feedback, and scene adaptability is enhanced through self-learning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of anti-collision warning systems, in particular to an anti-collision warning system for remote control operation of engineering machinery. BACKGROUND

[0002] Remote control operation of engineering machinery is widely used in dangerous scenes such as mines and construction, but its operation environment is complex, there are a large number of dynamic obstacles and equipment cross operation, and the collision risk is prominent. The traditional remote control system relies on the visual judgment of the operator, which is limited by the field of view of the camera and the delay, and it is difficult to fully perceive the surrounding environment, especially in poor conditions such as low light and high dust, the target recognition accuracy is greatly reduced, which is easy to cause safety accidents. The existing anti-collision system mainly uses single sensor detection, the laser radar has poor penetration in rain and fog, and the millimeter wave radar has insufficient identification of static targets, resulting in a blind area of perception or a high false alarm rate, which cannot meet the needs of complex scenes.

[0003] The risk assessment of the current warning system is mainly based on two-dimensional distance judgment, without considering the three-dimensional motion trajectory of the engineering machinery operation arm, which is difficult to accurately predict the collision risk. The warning decision lacks dynamic adaptability, and the priority of different targets such as personnel and equipment is not sufficient, the warning method is single, and the operator is easy to cause attention distraction and fail to respond in time. Remote communication relies on a single link, which is easy to be interrupted by shielding or electromagnetic interference, and the delay of control instructions may cause operation errors. The man-machine interface does not realize the intuitive presentation of risk information, and the operator needs to process a large amount of data, which increases the decision burden and affects the emergency response speed.

[0004] With the development of large-scale and intelligent engineering machinery, the safety demand of remote control operation is increasing. The existing technology has obvious shortcomings in multi-source perception fusion, dynamic risk assessment, intelligent warning decision, reliable communication guarantee and intuitive man-machine interaction, which leads to insufficient reliability and practicability of the anti-collision system. Collision accidents in complex operation scenes not only cause equipment damage, but also threaten the safety of operating personnel, which restricts the popularization and application of remote control technology. Therefore, it is necessary to develop an anti-collision warning system with all-around perception, accurate assessment, intelligent warning and reliable communication. SUMMARY

[0005] The anti-collision warning system for remote control operation of engineering machinery proposed in the present application solves the problems mentioned in the prior art.

[0006] In order to achieve the above purpose, the present application adopts the following technical scheme: an anti-collision warning system for remote control operation of engineering machinery, comprising: The perception module: eight sets of laser radars, four sets of millimeter wave radars and six sets of binocular cameras are deployed to form a full-range three-dimensional perception network; the horizontal field of view of the laser radar reaches 120 degrees, the vertical field of view is 30 degrees, and the point cloud data output density is 200 points per square meter; the millimeter wave radar detection coverage distance extends from 0.5 meters to 300 meters, and the speed measurement range is 0-100 km / h; the frame rate of the binocular camera is kept at 30 frames per second, the baseline distance is set to 120 mm, and data synchronization is realized through GPS timestamp; The dynamic target recognition module: YOLOv5 algorithm is used, and feature pyramid network structure is added to the original network to strengthen the detection ability of small targets in the working scene; the DBSCAN algorithm with density self-adaptation is applied to point cloud clustering, and the minimum number of neighbors required by the core point is dynamically adjusted between 8 and 15 according to the environment point cloud density; the system can identify excavators, dump trucks and other engineering machinery, accurately detect workers wearing reflective clothing, walls and static obstacles, and synchronously extract target position, speed vector and size parameters; The hazard situation assessment module: based on the extended Kalman filter algorithm, the target motion trajectory is predicted, and the safety state is evaluated by calculating the collision risk index, and the risk index calculation formula is , wherein d represents the relative distance in three-dimensional space, v is the modulus of the relative velocity vector, and θ is the included angle between the velocity directions; three levels are divided according to the risk index; The early warning decision module: a fuzzy control algorithm containing 25 rules is used, the input parameters include the quantized risk index with a value range of 0-10 seconds, the target priority is set in the order of personnel > equipment > obstacle, and the environmental complexity parameter reflecting the light condition and dust concentration.

[0007] Further, it also includes: The remote communication module: a 5G and WiFi6 dual-link redundant communication architecture is built, the 5G link uses independent networking mode, WiFi6 supports 802.11ax protocol, the dual-link state is monitored through heartbeat packets with an interval of 50 ms, and forward error correction coding is used in the data transmission process; The human-computer interaction module: the real-time picture console resolution is set to 2560x1440, the transparency of the early warning area is dynamically adjusted between 30%-80% according to the risk level to reflect the risk degree; the vibration frequency of the tactile feedback handle is adjusted in the range of 200-500 Hz, and the handle is equipped with a key to realize one-key switching of the early warning sensitivity mode; Sensor fusion calibration module: the coordinate conversion relationship between laser radar and millimeter wave radar is established through the chessboard calibration board, the conversion matrix includes translation X, Y, Z and rotation angle Roll, Pitch, Yaw parameters, least squares method is used for iterative optimization, so that the calibration residual is controlled within three centimeters, the system automatically executes the calibration process once a day.

[0008] Further, it also includes: Dynamic collision avoidance trajectory planning subsystem: when a high risk is detected, the RRT* algorithm is automatically started to plan the collision avoidance path, and the objective function of path optimization is Where a represents acceleration, j is jerk, d obs is the distance between the obstacle.

[0009] Further, the laser radar uses a 1550 nanometer wavelength laser light source, which has improved penetration rate in rain and fog weather compared to traditional wavelengths, and the scanning frequency is adjusted between 10 to 20 Hz according to the environmental complexity; the millimeter wave radar works in the 77 GHz frequency band, enhancing the anti-electromagnetic interference capability; the binocular camera is equipped with an infrared light supplement device, which automatically turns on when the light intensity is lower than 5 lux, and the image noise reduction applies the BM3D algorithm.

[0010] Further, the dynamic target recognition module has a deep learning model trained by samples built-in, the point cloud segmentation uses the Euclidean distance clustering algorithm, the clustering threshold is adaptively adjusted between 0.3 meters to 0.8 meters according to the environment, and the target tracking applies the Hungarian algorithm to achieve re-identification when the target is temporarily lost.

[0011] Further, the hazard situation assessment module adds the function of predicting the motion trajectory of the engineering machinery working arm, uses a cubic polynomial model p(t)=a0+a1t+a2t 2 +a3t 3 for prediction, where t is the time variable, and coefficients a0 to a3 are calculated by least squares method fitting the motion data of the past 500 milliseconds to realize the prediction of the end position of the working arm.

[0012] Further, the pre-warning decision module sets an emergency override mechanism, when the operator continuously presses the red emergency button for more than two seconds, the system releases the control lock in the red pre-warning state, and automatically records the operation time, device position and current risk value information.

[0013] Further, the remote communication module uses a data fragmentation transmission mechanism, the 5G link automatically switches to QPSK modulation mode when the received signal signal-to-noise ratio is lower than 10 decibels; WiFi6 supports Mesh networking mode, and control instructions are transmitted using AES-256 encryption algorithm.

[0014] Furthermore, the AR interface of the human-computer interaction module displays the target's movement direction through vector arrows. The arrow length is proportional to the target's speed, and the target's historical trajectory is retained for five seconds. The transparency of the trajectory lines gradually decreases over time. The haptic feedback handle is equipped with a directional prompt function. Vibration on the left indicates a risk on the left, and vibration on the right indicates a risk on the right. It also supports voice control, allowing users to adjust the display mode and warning parameters via voice.

[0015] Furthermore, it also includes: System self-learning optimization module: Based on the PPO deep reinforcement learning algorithm, a decision model is built. The system automatically analyzes the operation log every 24 hours to optimize the LiDAR scanning frequency sensor parameters and risk level classification thresholds. The model training adopts a federated learning framework to achieve joint training of data from various scenarios while protecting data privacy.

[0016] Compared with existing technologies, the beneficial effects of this invention are: The multi-source heterogeneous sensing module integrates the advantages of lidar, millimeter-wave radar and binocular camera to achieve all-weather, all-round environmental perception, eliminate the perception blind spots of a single sensor, and can still stably identify various targets in harsh environments, providing comprehensive and accurate environmental data for risk assessment.

[0017] Dynamic target recognition and risk assessment technology overcomes the limitations of traditional two-dimensional judgment. By combining three-dimensional trajectory prediction and target priority classification, it accurately identifies high-risk scenarios, avoiding false alarms and missed alarms. The early warning decision system dynamically adjusts the early warning method according to the risk level, forming a complete protection chain from visual prompts to proactive intervention. This ensures that operators can respond promptly and effectively avoid risks, reducing accidents caused by human error.

[0018] The dual-link redundant communication architecture solves the reliability problem of a single link. Through intelligent switching and data error correction technology, it ensures the real-time transmission of control commands and sensing data, avoiding the risk of loss of control caused by communication interruptions. AR augmented reality and human-computer interaction technology present complex risk information intuitively, and tactile feedback enhances early warning perception, reduces the operator's decision-making burden, and improves emergency response speed.

[0019] The system's self-learning and optimization capabilities enable it to adapt to different operating scenarios. By continuously optimizing sensor parameters and early warning strategies, it constantly improves recognition accuracy and timely warning. Multi-sensor fusion calibration and dynamic collision avoidance trajectory planning further ensure the system's stability and practicality. Overall, it realizes an upgrade from passive obstacle avoidance to active prevention in remote control operations, providing strong support for the safe and efficient operation of construction machinery. Attached Figure Description

[0020] Figure 1 This is a schematic block diagram of the anti-collision early warning system for remote control operation of engineering machinery proposed in this invention; Figure 2 This is a schematic diagram comparing the target recognition accuracy of different sensors in the anti-collision warning system for remote control operation of engineering machinery proposed in this invention. Figure 3 This is a schematic diagram comparing the self-learning optimization effect of the anti-collision warning system for remote control operation of engineering machinery proposed in this invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0023] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0024] Reference Figures 1 to 3 A collision avoidance warning system for remote-controlled operation of construction machinery, comprising: The perception module comprises eight LiDAR units, four millimeter-wave radar units, and six binocular cameras, forming a comprehensive three-dimensional perception network. The LiDAR units have a horizontal field of view of 120 degrees and a vertical field of view of 30 degrees, with a point cloud data output density of no less than 200 points per square meter, ensuring complete capture of environmental details. The millimeter-wave radar's detection range extends from 0.5 meters to 300 meters, and its speed measurement range covers 0 to 100 kilometers per hour, accurately capturing speed changes of moving targets. The binocular cameras maintain a frame rate of 30 frames per second, with a baseline distance set at 120 millimeters. Data synchronization is achieved through GPS timestamps, with a synchronization error controlled within 5 milliseconds, ensuring spatiotemporal consistency of data from different sensors.

[0025] Dynamic target recognition module: Employing an improved YOLOv5 algorithm, this module adds a feature pyramid network structure to the original network, enhancing the detection capability for small targets in the work environment. Point cloud clustering utilizes the density-adaptive DBSCAN algorithm, dynamically adjusting the minimum number of neighbors required for core points between eight and fifteen based on the environmental point cloud density. The system can identify six types of construction machinery, including excavators and dump trucks, in real time, accurately detect workers wearing reflective vests, and identify static obstacles such as walls and material piles. It simultaneously extracts target position, velocity vectors, and size parameters, ensuring positional accuracy within ten centimeters, velocity error not exceeding 0.2 meters per second, and size measurement error controlled below five percent.

[0026] Hazard assessment module: Predicts target trajectory based on extended Kalman filter algorithm, assesses safety status by calculating collision risk index, the formula for which is as follows: Where d represents the relative distance in three-dimensional space, v is the magnitude of the relative velocity vector, and θ is the angle between the velocity directions. Based on the risk index, three levels are defined: a risk index greater than five seconds indicates mild risk, a risk index between three and five seconds indicates moderate risk, and a risk index less than or equal to three seconds indicates high risk.

[0027] The early warning decision-making module employs a fuzzy control algorithm with twenty-five rules. Input parameters include a quantified risk index ranging from zero to ten seconds, target priority set in the order of personnel over equipment, equipment over obstacles, and environmental complexity parameters reflecting lighting conditions and dust concentration. A yellow warning displays the risk area in red via an AR interface; an orange warning adds an 80-decibel audible alert to the visual cues; and a red warning triggers a deceleration action of 0.5 meters per second squared and automatically locks the control lever in the dangerous direction to prevent accidental operation.

[0028] Remote communication module: A dual-link redundant communication architecture combining 5G and WiFi 6 is constructed. The 5G link adopts an independent networking mode with an uplink bandwidth of 50 Mbps, while WiFi 6 supports the 802.11ax protocol with a maximum transmission rate of 1.2 Gbps. The dual links monitor status via heartbeat packets spaced 50 milliseconds apart. Forward error correction coding is used during data transmission, with error correction capability not exceeding three bits per packet, ensuring command response latency is controlled within 15 milliseconds and guaranteeing the real-time performance of remote control.

[0029] Human-Computer Interaction Module: The AR control panel resolution is set to 2560 x 1440. The transparency of the warning area is dynamically adjusted between 30% and 80% according to the risk level, intuitively reflecting the degree of risk. The vibration frequency of the haptic feedback handle is adjustable within the range of 200 to 500 Hz, with 50 Hz intervals. The vibration intensity is divided into five levels, corresponding to a force of two to eight Newtons. The handle is equipped with a dedicated button that allows for one-click switching of the warning sensitivity mode to adapt to the usage habits of different operators.

[0030] This invention also includes: Sensor Fusion Calibration Subsystem: The coordinate transformation relationship between lidar and millimeter-wave radar is established through a checkerboard calibration board. The transformation matrix includes translation parameters X, Y, Z and rotation parameters Roll, Pitch, and Yaw. Iterative optimization is performed using the least squares method to keep the calibration residual within three centimeters. The system automatically performs a calibration process once a day to ensure spatial consistency of data from different sensors.

[0031] Dynamic collision avoidance trajectory planning subsystem: When a high risk is detected, the RRT* algorithm is automatically activated to plan the collision avoidance path. The objective function for path optimization is... Where 'a' represents acceleration, with a maximum value not exceeding one square meter per second, 'j' represents jerk, with a maximum value controlled within five cubic meters per second, and 'd' represents acceleration. obs To ensure the smooth movement of the robotic arm, a safety margin of 1.5 meters is set between the arm and obstacles, and the curvature of the planned trajectory does not exceed 0.03 per meter.

[0032] In this invention, the lidar uses a laser light source with a wavelength of 1,550 nanometers, which improves the penetration rate by 30% in rainy and foggy weather compared to traditional wavelengths. The scanning frequency can be adjusted between 10 and 20 Hz according to the complexity of the environment. The millimeter-wave radar operates in the 77 GHz band, enhancing its anti-electromagnetic interference capability and achieving a distance resolution of 0.2 meters. The binocular camera is equipped with an infrared supplementary light device that automatically turns on when the light intensity is below 5 lux. Image noise reduction uses the BM3D algorithm, which improves the signal-to-noise ratio by more than 15 dB. The dynamic target recognition module incorporates a deep learning model trained on over 10,000 samples, maintaining an accuracy rate of over 98% in detecting individuals wearing reflective clothing, and can reliably identify targets even when the occlusion level is less than 30%. Point cloud segmentation employs an Euclidean distance clustering algorithm, with the clustering threshold adaptively adjusted between 0.3 and 0.8 meters based on the environment. Target tracking utilizes the Hungarian algorithm, enabling re-identification within one second if the target is briefly lost, ensuring continuous tracking.

[0033] In this invention, the hazard assessment module adds a function to predict the trajectory of the boom of construction machinery, using a cubic polynomial model p(t) = a0 + a1t + a2t. 2 +a3t 3 The prediction is performed, where t is the time variable, and the coefficients a0 to a3 are calculated by fitting the motion data of the past 500 milliseconds using the least squares method to achieve accurate prediction of the position of the end of the working arm. The prediction error is controlled within 8 centimeters with a confidence level of 90%. The early warning decision module is equipped with an emergency override mechanism. When the operator presses the red emergency button for more than two seconds, the system can release the control lock under the red warning state. At the same time, it automatically records key information such as operation time, equipment location and current risk value. After unlocking, the system maintains the original warning prompt for ten seconds to prevent the operator from causing safety hazards due to negligence.

[0034] In this invention, the remote communication module adopts a data fragmentation transmission mechanism, with each packet having a maximum data length of 1,500 bytes. The 5G link automatically switches to QPSK modulation when the received signal-to-noise ratio is below 1 / 2 decibel, ensuring communication quality in weak signal environments. WiFi 6 supports Mesh networking mode, allowing for the expansion of up to eight nodes to enhance coverage. Key control commands are transmitted using AES-256 encryption to prevent unauthorized interception or tampering of data. The AR interface of the human-computer interaction module displays the target's movement direction through vector arrows. The arrow length is proportional to the target's speed, and the target's historical trajectory is retained for five seconds, with the transparency of the trajectory lines gradually decreasing over time. The haptic feedback handle has a directional indication function; vibration on the left indicates a risk on the left, and vibration on the right indicates a risk on the right. It supports voice control, with a voice command recognition rate of over 95%, and the display mode and warning parameters can be adjusted via voice.

[0035] This invention also includes: The system's self-learning optimization subsystem constructs an intelligent decision-making model based on the PPO deep reinforcement learning algorithm. The reward function is a warning accuracy rate of no less than 90% and a false alarm rate of no more than 5%. The system automatically analyzes operation logs every 24 hours to optimize sensor parameters such as LiDAR scanning frequency and risk level classification thresholds. Model training employs a federated learning framework, enabling joint training of multi-scenario data while protecting data privacy.

[0036] Specific implementation method of collision avoidance warning system for remote control operation of construction machinery.

[0037] This embodiment takes the remote control operation of a mining excavator as an example to explain in detail the implementation process of the anti-collision warning system. The system is compatible with 20-50 ton excavators and covers typical scenarios such as mining and material loading and unloading.

[0038] I. Multi-source heterogeneous sensing module The excavator is equipped with eight LiDAR systems deployed circumferentially around its body, featuring a 120° horizontal field of view and a 30° vertical field of view. Adjacent LiDARs overlap by 20° to eliminate blind spots, achieving a point cloud output density of 200 points / m². A 1550nm wavelength laser enhances rain and fog penetration. Four millimeter-wave radars, operating in the 77GHz band, are installed at the four corners of the excavator, with a detection range of 0.5-300m, a speed range of 0-100km / h, and a range resolution of 0.2m. Six binocular cameras are distributed around the cab, operating at 30fps with a baseline distance of 120mm. Built-in infrared illuminators automatically activate when illumination is ≤5 lux, and BM3D algorithm noise reduction improves image quality in low-light environments. Sensor synchronization is achieved via GPS timestamps, with a synchronization error ≤5ms, ensuring data spatiotemporal alignment.

[0039] II. Dynamic Target Recognition and Feature Extraction The improved YOLOv5 algorithm incorporates a feature pyramid network to enhance detection of small targets (such as workers at a distance). The model was trained on over 10,000 samples, covering six types of construction machinery, personnel in reflective clothing, and five types of static obstacles. Point cloud clustering employs the density-adaptive DBSCAN algorithm, with the minimum neighbor number for core points dynamically adjusted between 8 and 15 based on point cloud density. The Euclidean distance clustering threshold adaptively varies from 0.3 to 0.8 meters. Real-time extraction of target 3D position (accuracy ±10cm), velocity vector (error ≤0.2m / s), and size parameters (error ≤5%) is performed to construct a dynamic target feature database, supporting continuous tracking of target trajectories. Lost targets can be re-identified within 1 second.

[0040] III. Risk Assessment and Early Warning Decision-Making The collision risk index is calculated based on the extended Kalman filter to predict the target's trajectory in the next 3 seconds. Where d is the relative distance in three-dimensional space, v is the relative velocity magnitude, and θ is the angle between the velocity directions. Risk levels are divided as follows: R > 5s (mild), 3s < R ≤ 5s (moderate), and R ≤ 3s (high). The fuzzy control algorithm contains 25 rules, with input variables quantified as risk index [0,10]s, target priority (personnel > equipment > obstacles), and environmental complexity (light / dust level). Yellow alerts are marked with red risk areas on the AR interface, orange alerts are superimposed with an 80dB buzzer, and red alerts are triggered at 0.5m / s. 2 Decelerate and lock the dangerous direction control lever, while retaining the emergency override mechanism (press and hold the red button for 2 seconds to unlock).

[0041] IV. Remote Communication and Human-Computer Interaction The system achieves dual-link communication between 5G SA architecture and WiFi 6, with a 5G uplink bandwidth of 50Mbps and WiFi 6 supporting the 802.11ax protocol (1.2Gbps). Link status is monitored via 50ms heartbeat packets, employing forward error correction coding (error correction ≤ 3 bits / packet), and command latency ≤ 15ms. The AR control panel has a resolution of 2560×1440, with dynamic adjustment of the warning area transparency from 30% to 80%, displaying a target motion vector arrow (length proportional to speed) and a 5-second historical trajectory (fading effect). The haptic handle vibrates at a frequency of 200-500Hz with intensity levels of 2-8N; left-side vibration indicates left-side risk, and voice control (recognition rate ≥ 95%) is supported for switching display modes.

[0042] V. Dynamic Collision Avoidance and Self-Learning Optimization Initiate RRT* trajectory planning and optimize the objective function when facing high risk. Acceleration limit ≤ 1 m / s² 2 Acceleration ≤ 5m / s 3 The obstacle avoidance safety margin is 1.5m. The self-learning subsystem is based on the PPO algorithm, aiming for a warning accuracy of ≥90% and a false alarm rate of ≤5%. It analyzes logs 24 hours a day to optimize sensor parameters (such as LiDAR scanning frequency) and trains the model using federated learning and multi-scenario data.

[0043] Implementation results data: Evaluation metrics Conventional system System of the present application Target recognition accuracy Lower Higher Collision risk prediction accuracy Generally Higher Communication interruption rate Higher Lower Operator emergency response time Longer Shorter Complex environment adaptability Worse Better Data comparison validates the comprehensive advantages of the system presented in this application. Improved target recognition accuracy is attributed to multi-source sensor fusion and algorithm optimization, eliminating blind spots from single sensors and maintaining stable recognition capabilities even in harsh environments. Increased collision risk prediction accuracy stems from 3D trajectory prediction and dynamic risk assessment, accurately distinguishing different risk levels and reducing false alarms and missed alarms. The dual-link communication architecture significantly reduces the interruption rate, and forward error correction coding ensures reliable data transmission, avoiding operational delays caused by communication problems. Intuitive AR display and haptic feedback reduce operator workload, shorten emergency response time, and improve hazard handling efficiency. Self-learning optimization capabilities enable the system to adapt to different scenarios, significantly enhancing stability and adaptability in complex environments, and overall improving the safety and reliability of remote control operations.

[0044] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A collision avoidance warning system for remote control operation of engineering machinery, characterized in that, include: The perception module consists of eight LiDAR units, four millimeter-wave radar units, and six binocular cameras, forming an all-around three-dimensional perception network. The LiDAR units have a horizontal field of view of 120 degrees and a vertical field of view of 30 degrees, with a point cloud data output density of 200 points per square meter. The millimeter-wave radars extend their detection range from 0.5 meters to 300 meters and have a speed measurement range of 0 to 100 kilometers per hour. The binocular cameras maintain a frame rate of 30 frames per second, with a baseline distance set to 120 millimeters, and data synchronization is achieved through GPS timestamps. Dynamic target recognition module: Adopting the YOLOv5 algorithm, a feature pyramid network structure is added to the original network to enhance the detection capability of small targets in the work scene; the point cloud clustering uses the density-adaptive DBSCAN algorithm, and the minimum number of neighbors required for the core point is dynamically adjusted between eight and fifteen according to the density of the environmental point cloud; the system can identify excavators and dump trucks in real time, accurately detect workers wearing reflective vests, walls, and static obstacles such as stacked materials, and simultaneously extract the target position, velocity vector and size parameters; Hazard assessment module: Predicts target trajectory based on extended Kalman filter algorithm, assesses safety status by calculating collision risk index, the formula for which is as follows: Where d represents the relative distance in three-dimensional space, v is the magnitude of the relative velocity vector, and θ is the angle between the velocity directions; three levels are divided according to the risk index; Early warning decision module: It adopts a fuzzy control algorithm with 25 rules. The input parameters include a quantified risk index with a value range of 0 to 10 seconds, target priority set in the order of personnel above equipment, equipment above obstacles, and environmental complexity parameters reflecting lighting conditions and dust concentration.

2. The anti-collision warning system for remote control operation of engineering machinery according to claim 1, characterized in that, Also includes: Remote communication module: Construct a dual-link redundant communication architecture of 5G and WiFi 6. The 5G link adopts an independent networking mode, and WiFi 6 supports the 802.11ax protocol. The dual links monitor the status through heartbeat packets at 50 millisecond intervals, and forward error correction coding is used during data transmission. Human-computer interaction module: The resolution of the real-time screen control panel is set to 2,560 x 1,440. The transparency of the warning area is dynamically adjusted between 30% and 80% according to the risk level to reflect the degree of risk. The vibration frequency of the haptic feedback handle can be adjusted within the range of 200 to 500 Hz. The handle is equipped with buttons to achieve one-click switching of the warning sensitivity mode. Sensor fusion calibration module: The coordinate transformation relationship between lidar and millimeter-wave radar is established through a checkerboard calibration board. The transformation matrix includes translation parameters X, Y, Z and rotation parameters Roll, Pitch, Yaw. Iterative optimization is performed using the least squares method to control the calibration residual within three centimeters. The system automatically performs the calibration process once a day.

3. The anti-collision warning system for remote control operation of engineering machinery according to claim 1, characterized in that, Also includes: Dynamic collision avoidance trajectory planning subsystem: When a high risk is detected, the RRT* algorithm is automatically activated to plan the collision avoidance path. The objective function for path optimization is... Where a represents acceleration, j represents jerk, and d represents acceleration. obs This represents the distance to the obstacle.

4. The anti-collision warning system for remote control operation of engineering machinery according to claim 1, characterized in that, The lidar uses a 1,550-nanometer wavelength laser light source, which improves the penetration rate in rainy and foggy weather compared to traditional wavelengths. The scanning frequency is adjusted between 10 and 20 Hz according to the complexity of the environment. The millimeter-wave radar operates in the 77 GHz band, enhancing its anti-electromagnetic interference capability. The binocular camera is equipped with an infrared fill light device that automatically turns on when the light intensity is below 5 lux. Image noise reduction uses the BM3D algorithm.

5. The anti-collision warning system for remote control operation of engineering machinery according to claim 1, characterized in that, The dynamic target recognition module incorporates a deep learning model trained on samples. Point cloud segmentation uses the Euclidean distance clustering algorithm, and the clustering threshold is adaptively adjusted between 0.3 meters and 0.8 meters according to the environment. Target tracking uses the Hungarian algorithm, enabling re-identification when the target is briefly lost.

6. The anti-collision early warning system for remote control operation of engineering machinery according to claim 1, characterized in that, The newly added hazard assessment module includes a function to predict the trajectory of the boom of construction machinery, using a cubic polynomial model p(t) = a0 + a1t + a2t. 2 +a3t 3 A prediction is made, where t is the time variable, and the coefficients a0 to a3 are calculated by fitting the motion data of the past 500 milliseconds using the least squares method to predict the position of the end effector of the working arm.

7. The anti-collision warning system for remote control operation of engineering machinery according to claim 1, characterized in that, The early warning decision module is equipped with an emergency override mechanism. When the operator presses the red emergency button for more than two seconds, the system will release the control lock under the red early warning state and automatically record the operation time, equipment location and current risk value information.

8. The anti-collision warning system for remote control operation of engineering machinery according to claim 2, characterized in that, The remote communication module adopts a data fragmentation transmission mechanism. When the signal-to-noise ratio of the received signal is less than ten decibels, the 5G link automatically switches to QPSK modulation mode. WiFi6 supports Mesh networking mode, and control commands are transmitted using the AES-256 encryption algorithm.

9. The anti-collision early warning system for remote control operation of engineering machinery according to claim 2, characterized in that, The AR interface of the human-computer interaction module displays the target's movement direction through vector arrows. The arrow length is proportional to the target's speed. The target's historical trajectory is retained for five seconds, and the transparency of the trajectory lines gradually decreases over time. The haptic feedback handle features directional indicators: vibration on the left indicates a risk on the left, and vibration on the right indicates a risk on the right. It also supports voice control, allowing users to adjust the display mode and warning parameters via voice commands.

10. The anti-collision warning system for remote control operation of engineering machinery according to claim 1, characterized in that, Also includes: The system's self-learning optimization module: Based on the PPO deep reinforcement learning algorithm, a decision model is built. The system automatically analyzes the operation logs every 24 hours to optimize the LiDAR scanning frequency sensor parameters and risk level classification thresholds. The model training adopts a federated learning framework, which enables joint training of data from various scenarios while protecting data privacy.

Images