Operation safety distance management method based on laser radar point cloud data

Abstract

The invention relates to the technical field of safety monitoring, and particularly discloses an operation safety distance management method based on laser radar point cloud data, and the method comprises the steps: generating a dangerous point cloud set based on an initial three-dimensional point cloud model of a complete field environment; performing coordinate transformation on the point cloud data currently collected by the three-dimensional laser radar equipment based on the rigid body transformation matrix to obtain current point cloud data; registering the current point cloud data with the initial three-dimensional point cloud model and identifying all independent dynamic targets; determining a danger approaching trend of each independent dynamic target based on the motion prediction trajectory of each independent dynamic target; determining a dynamic safety distance threshold value of each dangerous dynamic target based on the danger approaching trend and the response delay of each independent dynamic target and the additional safety distance of the corresponding dangerous source area; triggering a corresponding alarm mechanism based on the dynamic safety distance threshold and the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set; and the overall safety of a working site is improved.

Description

Technical Field

[0001] The present invention relates to the field of safety monitoring technology, and in particular to an operation safety distance management method based on laser radar point cloud data. Background Art

[0002] With the continuous advancement of industrial automation, a large number of potential hazards, such as high-temperature and high-pressure equipment, high-speed moving machinery, and live electrical devices, exist in power plants, substations, factories, workshops, and warehouses and logistics centers. To ensure personnel safety and prevent personal injury from accidentally entering or approaching hazardous areas, safety protection systems based on sensor technology have gained widespread attention and application in recent years. Currently, common hazard detection technologies include infrared beam sensors, ultrasonic ranging, two-dimensional lidar (LiDAR), video image recognition, and single-type point cloud analysis. For example, infrared sensors use the principle of transmitting and receiving light beams to determine whether an object has entered a preset area; ultrasonic sensors use the principle of sound wave reflection to measure distance and are suitable for short-range obstacle detection; and two-dimensional LiDAR scans plane contours and combines edge detection algorithms to provide approach warnings. Alternatively, cameras combined with image recognition algorithms can be used to determine proximity to a hazardous area based on human posture and the overlap of the human position area with the hazardous area. These traditional technologies commonly suffer from the following issues in practical applications: Limited spatial perception capabilities: Infrared and ultrasonic sensors only provide one- or two-dimensional distance information, making it difficult to construct a complete three-dimensional environmental model and accurately determine the spatial position and movement trajectory of an object. Susceptibility to environmental interference: Image recognition relies on lighting conditions and is prone to misjudgment in complex backgrounds. Ultrasonic sensors are susceptible to environmental factors such as temperature, humidity, and wind speed. Lack of dynamic adaptability: Most systems are fixed deployments. Once installed, if the equipment moves or rotates, the monitoring baseline cannot be adjusted in a timely manner, resulting in increased false alarm rates. Poor adaptability to unstructured scenes: In industrial sites with complex backgrounds, such as those with large numbers of mobile equipment and temporarily stored objects, conventional systems struggle to effectively distinguish between background and target, leading to missed or false detections. Delayed alarm response: Due to the lack of effective trajectory prediction mechanisms, existing systems often use static threshold judgment methods, which are unable to predict dangerous trends in advance and reduce the system's active defense capabilities. In recent years, the development of 3D LiDAR technology has demonstrated powerful spatial modeling and target recognition capabilities in fields such as robot navigation, autonomous driving, and security monitoring. 3D LiDAR can acquire high-density, full-viewpoint point cloud data, providing enhanced environmental perception and interference resistance. Existing research has attempted to apply 3D LiDAR to industrial safety, using point cloud processing such as clustering, segmentation, and classification to locate and track people or objects.

[0003] Although existing technologies have achieved the detection of people or objects near dangerous sources to a certain extent, they still have obvious shortcomings in the following aspects: when the radar equipment is displaced due to vibration, environmental factors or equipment displacement, the point cloud model is prone to mismatch, resulting in target recognition errors and distance calculation deviations, thereby causing false alarms or missed alarms. Therefore, there is a lack of effective compensation mechanism for changes in equipment posture; the time and spatial information of three-dimensional point cloud data is not fully utilized, and there is a lack of prediction of target movement trends; there is also a lack of accurate determination of the dynamic safety distance threshold for each dangerous dynamic target, which leads to the need to improve the system's response speed and ability to prevent sudden intrusion incidents.

[0004] Therefore, the present invention proposes an operation safety distance management method based on lidar point cloud data. Summary of the Invention

[0005] This invention provides a method for managing safe operating distances based on LiDAR point cloud data. It aims to address at least some of the technical deficiencies found in existing hazardous source monitoring systems, including model mismatches caused by equipment position changes, severe background interference in point clouds, low dynamic target recognition accuracy, delayed safe distance determination, and high false alarm rates. By integrating a multimodal data fusion mechanism with 3D LiDAR and a nine-axis sensor, combined with point cloud processing algorithms and trajectory prediction methods, it achieves high-precision, real-time, and intelligent monitoring of people and objects approaching hazardous sources in industrial and other environments.

[0006] The present invention provides a method for managing safe distances for operations based on laser radar point cloud data, comprising:

[0007] Generate a collection of all hazardous point clouds based on the initial 3D point cloud model of the complete scene environment;

[0008] Perform coordinate transformation on the point cloud data currently collected by the 3D laser radar device based on the rigid body transformation matrix to obtain the current point cloud data;

[0009] Align the current point cloud data with the initial 3D point cloud model and identify all independent dynamic targets;

[0010] Determine the dangerous approach trend of each independent dynamic target based on the predicted motion trajectory of each independent dynamic target;

[0011] Determine a dynamic safety distance threshold for each dangerous dynamic target based on the dangerous approach trend and response delay of each independent dynamic target and the additional safety distance of the corresponding dangerous source area;

[0012] The corresponding alarm mechanism is triggered based on the dynamic safety distance threshold of each dangerous dynamic target and the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set.

[0013] Preferably, a set of all hazardous point clouds is generated based on an initial 3D point cloud model of the complete scene environment, including:

[0014] Use 3D LiDAR scanning to integrate the on-site environment and generate an initial 3D point cloud model;

[0015] All hazard source areas are selected in the initial three-dimensional point cloud model, and the point cloud ranges of all hazard source areas are marked to form a set of all hazardous point clouds.

[0016] Preferably, the process of obtaining the rigid body transformation matrix includes:

[0017] The attitude angle and translation of the three-dimensional laser radar device are detected in real time based on the nine-axis sensor, and the rigid body transformation matrix of the three-dimensional laser radar device is constructed based on the attitude angle and translation of the three-dimensional laser radar device.

[0018] Preferably, the current point cloud data is registered with the initial 3D point cloud model and all independent dynamic targets are identified, including:

[0019] Register the current point cloud data with the initial three-dimensional point cloud model to obtain the point cloud registration result;

[0020] Separate all dynamic target points in the current point cloud data based on the point cloud registration results;

[0021] Perform cluster analysis on all dynamic target points and identify all independent dynamic targets.

[0022] Preferably, cluster analysis is performed on all dynamic target points to identify all independent dynamic targets, including:

[0023] Determine a neighborhood point cloud of each dynamic target point in the current point cloud data, and determine a local point cloud density of each dynamic target point based on the neighborhood point cloud of each dynamic target point;

[0024] An adaptive density threshold is calculated based on the mean and standard deviation of the local point cloud density of all dynamic target points and a preset standard deviation coefficient;

[0025] All dynamic target points whose local point cloud density is not less than the adaptive density threshold are regarded as all core target points;

[0026] Determine the spatial motion joint distance metric data based on the current point cloud and the point cloud at the previous acquisition moment for each core target point;

[0027] Based on the spatial motion joint distance metric data and the local point cloud density of all core target points, multiple core target points are determined as all cluster centers;

[0028] All dynamic target points except all cluster centers are assigned to the cluster center whose local point cloud density is greater than the local point cloud density of the corresponding dynamic target point and is closest to the cluster center, obtaining multiple point clusters, and treating each point cluster as a single independent dynamic target.

[0029] Preferably, determining spatial motion joint distance metric data based on each core target point in the current point cloud and the point cloud at the previous acquisition moment includes:

[0030] Determine the three-dimensional coordinates of each core target point in the current point cloud and the point cloud acquired at the previous acquisition moment;

[0031] The vectors pointing from the origin of the preset three-dimensional coordinate system to the three-dimensional coordinates of each core target point in the current point cloud and the point cloud obtained at the previous acquisition moment are regarded as the three-dimensional position vectors of each core target point in the current point cloud and the point cloud obtained at the previous acquisition moment;

[0032] The 2-norm of the difference between the three-dimensional position vectors of each two core target points in the current point cloud is used as the spatial metric value corresponding to the two core target points;

[0033] The difference between the three-dimensional position vectors of each core target point in the current point cloud and the point cloud acquired at the previous acquisition moment is regarded as the inter-frame velocity vector of each core target point;

[0034] The 2-norm of the difference between the velocity vectors of each two core target points is used as the motion distance measurement value corresponding to the two core target points;

[0035] The spatial motion joint distance measurement data includes the spatial measurement values ​​and motion distance measurement values ​​of all pairwise core target points.

[0036] Preferably, multiple core target points are determined as all cluster centers based on the spatial motion joint distance metric data and the local point cloud density of all core target points, including:

[0037] Determine the defined distance between each two core target points based on the spatial metric value and the motion distance metric value of each two core target points in the spatial motion joint distance metric data;

[0038] Among all core target points, all core target points with a local point cloud density greater than that of a single core target point are selected as a set of high-density candidate target points corresponding to the single core target point.

[0039] The minimum value of the defined distances between each core target point and the corresponding core target point in the large-density candidate target point set of each core target point is taken as the decision distance of the corresponding core target point;

[0040] Based on the decision distance of all core target points, multiple core target points are screened out from all core target points as all cluster centers.

[0041] Preferably, the process of obtaining the motion prediction trajectory of each independent dynamic target includes:

[0042] A point cloud data set sequence of each independent dynamic target within the historical period is intercepted from the point cloud data of the complete field environment acquired within the historical period, and a historical motion trajectory of each independent dynamic target is fitted based on the point cloud data set sequence of each independent dynamic target within the historical period;

[0043] A Kalman filter is introduced, and the motion trajectory of each independent dynamic target is predicted based on the historical motion trajectory of each independent dynamic target to determine the motion prediction trajectory of each independent dynamic target.

[0044] Preferably, determining the dynamic safety distance threshold of each dangerous dynamic target based on the dangerous approach trend and response delay of each independent dynamic target and the additional safety distance of the corresponding danger source area includes:

[0045] Based on the dangerous approach trends of all independent dynamic targets, all dangerous dynamic targets are screened out from all independent dynamic targets;

[0046] Determine the dangerous approach speed of each dangerous dynamic target based on the predicted motion trajectory of each dangerous dynamic target;

[0047] The sum of the product of the safety device response time, the control system response time and the system delay, the dangerous approach speed of each dangerous dynamic target and the additional safety distance of the corresponding hazard source area is taken as the dynamic safety distance threshold of the corresponding dangerous dynamic target.

[0048] Preferably, triggering a corresponding alarm mechanism based on the dynamic safety distance threshold of each dangerous dynamic target and the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set includes:

[0049] When the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set is less than 1 times the dynamic safety distance threshold of the corresponding dangerous dynamic target, an audible and visual alarm is triggered;

[0050] When the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set is less than 0.7 times the dynamic safety distance threshold of the corresponding dangerous dynamic target, the linkage control is triggered to limit the equipment action in the corresponding dangerous source area;

[0051] When the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set is less than 0.5 times the dynamic safety distance threshold of the corresponding dangerous dynamic target, the equipment in the corresponding dangerous source area is triggered to shut down urgently and protective measures are started.

[0052] The present invention offers the following advantages over existing technologies: By generating a collection of hazardous point clouds based on an initial 3D point cloud model of the complete on-site environment, it can clearly define hazardous areas on the worksite, providing a clear reference standard for subsequent safety distance management and enabling more targeted hazard identification. The current point cloud data collected by the 3D lidar device is transformed using a rigid body transformation matrix to ensure data processing within a unified coordinate system, improving data accuracy and consistency and facilitating subsequent precise analysis. The current point cloud data is aligned with the initial 3D point cloud model and all independent dynamic targets are identified, providing real-time visibility into the dynamic targets on-site, laying the foundation for further risk assessment. The predicted motion trajectory of each independent dynamic target determines its approach trend, enabling early prediction of the target's proximity to the hazardous area and providing a proactive warning of potential hazards. The dynamic safety distance threshold for each dangerous dynamic target is determined based on the approach trend, response delay, and the additional safety distance from the hazard source area. This takes into account multiple practical factors, making the safety distance setting more tailored to the actual work scenario and enhancing the effectiveness of safety management. The corresponding alarm mechanism is triggered according to the dynamic safety distance threshold and the distance between the independent dynamic target point cloud set and the dangerous point cloud set, so as to achieve timely response to danger, maximize the safety of workers and equipment, reduce the probability of safety accidents, and improve the overall safety of the work site.

[0053] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in this application document.

[0054] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0056] Figure 1 This is a flow chart of a method for managing safe distances for operations based on lidar point cloud data in an embodiment of the present invention;

[0057] Figure 2 Schematic diagram of an independent dynamic target recognition process in an embodiment of the present invention;

[0058] Figure 3 Schematic diagram of the process of determining spatial motion joint distance metric data in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0060] like Figure 1 As shown, the present invention provides a real-time method for operating safety distance management based on laser radar point cloud data, including:

[0061] Generate a collection of all hazardous point clouds based on the initial 3D point cloud model of the complete scene environment;

[0062] Based on the rigid body transformation matrix T, the point cloud data P currently collected by the 3D laser radar device is transformed into current Perform coordinate transformation to obtain the current point cloud data P′ current , including: P′ current =T*P current ;

[0063] Align the current point cloud data with the initial 3D point cloud model and identify all independent dynamic targets;

[0064] The dangerous approach trend of each independent dynamic target is determined based on the predicted motion trajectory of each independent dynamic target; the approach behavior is predicted in advance, so that the control system has enough time to respond, greatly reducing the risk of accidents.

[0065] Determining the dynamic safety distance threshold for each dangerous dynamic target based on the dangerous approach trend and response delay of each independent dynamic target and the additional safety distance of the corresponding hazard source area can overcome the problem that the existing system mostly uses static safety threshold determination method, lacks the ability to predict the target movement trend, causes alarm response lag, and makes it difficult to achieve active protection;

[0066] The corresponding alarm mechanism is triggered based on the dynamic safety distance threshold of each dangerous dynamic target and the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set (that is, the distance between the nearest point clouds or the minimum Euclidean distance, the value of which must comply with the standards ISO13855 and GB / T 41355-2022).

[0067] In this embodiment, the complete on-site environment refers to the entire spatial scope of the workplace and all objects and facilities contained therein. For example, in a factory workshop, the complete on-site environment includes all areas and objects within the workshop, such as machinery and equipment, material storage areas, passageways, and operating platforms.

[0068] In this embodiment, the point cloud data currently collected by the 3D LiDAR device refers to point cloud data acquired by the 3D LiDAR device during real-time scanning of the scene. This data reflects the spatial position information of objects in the scene at the current moment. For example, at a certain moment, the device may scan a worker operating a machine in a workshop and several moving transport vehicles. The position information of these objects is presented in the form of point cloud data.

[0069] In this embodiment, response delay includes the safety device response time, control system response time, and system delay. It refers to the time between triggering the safety mechanism and actually taking the response action after a dangerous situation is detected. For example, the time between the safety device detecting an object approaching a dangerous area and sending a signal and the control system receiving the signal and taking the corresponding action (such as shutting down the equipment) is the response delay.

[0070] In this embodiment, the hazard source area refers to a specific area at the work site where potential hazards exist, such as the high-voltage electrical area of ​​a power plant, the area around a high-temperature furnace in a factory, etc.

[0071] In this embodiment, the additional safety distance for the hazard zone is defined as a safety distance set in addition to the boundaries of the hazard zone itself. For example, for high-speed rotating machinery, an additional safety distance may be set in addition to the machine's outline to further ensure safety, preventing people or objects from accidentally approaching and being drawn into the area. This distance is determined based on factors such as the hazard level and the operational scenario.

[0072] In this embodiment, a dangerous dynamic target is an independent dynamic target at the work site that, based on its predicted trajectory, is likely to approach a hazardous area and could potentially cause a dangerous situation. For example, a forklift in a workshop, traveling toward a high-temperature furnace, is considered a dangerous dynamic target because its trajectory indicates it may enter a hazardous area.

[0073] In this embodiment, the dynamic safety distance threshold is a safety distance value determined for each dangerous dynamic target by comprehensively considering its dangerous approach tendency, response delay, and the additional safety distance to the corresponding hazard source area. For example, if a dangerous dynamic target approaches the danger zone of a high-temperature furnace at a certain speed, the dynamic safety distance threshold is determined to be 5 meters, taking into account the response delay and the additional safety distance to the furnace area.

[0074] The hardware structure of this method includes:

[0075] 3D LiDAR module: collects on-site spatial point cloud data in real time, builds a 3D environment model, and continuously obtains the location information of people or objects.

[0076] Nine-axis sensor module: includes a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer, used to detect changes in the radar device's posture (attitude angle and position) and provide an initial transformation matrix to assist in point cloud registration.

[0077] Main control processing unit (MCU / FPGA / embedded platform / CPU / GPU, etc.): responsible for the execution of core algorithms such as point cloud data preprocessing, background filtering, target extraction, distance calculation, trajectory prediction, and alarm logic.

[0078] Communication and alarm module: uploads the detection results to the monitoring center and triggers warning mechanisms such as sound and light alarms and remote push notifications when the proximity to the danger source reaches the set threshold.

[0079] Power management module: Provides stable power supply and supports multiple power supply methods (such as DC adapter, PoE, battery, etc.) to ensure long-term system operation.

[0080] Most existing systems are developed to meet specific application scenarios, resulting in relatively rigid functionality and limited flexibility in adapting to different types of hazards or deployment environments. This invention divides the system into multiple independent modules, including data acquisition, posture compensation, point cloud processing, target recognition, distance calculation, trajectory prediction, and alarm output. This facilitates integration and expansion based on diverse application needs. Furthermore, it supports remote configuration and parameter updates, enhancing the system's flexibility and applicability.

[0081] In order to precisely define the hazardous areas at the work site and provide accurate and targeted hazard references for subsequent safety distance management based on point cloud data, we further proposed generating a set of all hazardous point clouds based on the initial 3D point cloud model of the complete site environment, including:

[0082] Use 3D LiDAR scanning to integrate the on-site environment and generate an initial 3D point cloud model;

[0083] All hazard source areas are selected in the initial three-dimensional point cloud model, and the point cloud ranges of all hazard source areas are marked to form a set of all hazardous point clouds.

[0084] In this embodiment, all hazard source areas are selected from the initial 3D point cloud model: specific areas within the initial 3D point cloud model that pose potential hazards are identified. These areas are hazard source areas. For example, in a factory scenario, areas with high-voltage electrical equipment or high-speed machinery, which could pose a hazard to personnel or equipment, are the hazard source areas to be selected.

[0085] In this embodiment, all hazard source areas are marked by point cloud ranges to form a collection of all hazardous point clouds. After selecting the hazard source areas, the specific range boundaries of each hazard source area are determined and marked in the point cloud model. This marking method separates the point cloud data belonging to each hazard source area, and these marked point cloud data form the hazardous point cloud collection.

[0086] In order to achieve accurate recognition of dynamic targets on site, a process for obtaining the rigid body transformation matrix is ​​further proposed, including:

[0087] The attitude angle and translation of the three-dimensional laser radar device are detected in real time based on the nine-axis sensor, and the rigid body transformation matrix of the three-dimensional laser radar device is constructed based on the attitude angle and translation of the three-dimensional laser radar device.

[0088] In this embodiment, a nine-axis sensor is used to detect the attitude angle and translation of a 3D LiDAR device in real time. The nine-axis sensor integrates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer, enabling real-time monitoring of the 3D LiDAR device's state changes in space. The attitude angle describes the orientation of the 3D LiDAR device, such as whether it is horizontal, tilted at a certain angle, or rotated. The nine-axis sensor can accurately capture this angle information. The translation indicates the movement of the 3D LiDAR device in three-dimensional space, such as the distance it has moved in the x, y, and z axes. In practical application scenarios, such as in a continuously operating industrial monitoring environment, the attitude and position of the 3D LiDAR device may change due to external factors (such as device vibration and platform movement). The nine-axis sensor can capture these changes in real time. Suppose the 3D LiDAR device is installed on a mobile robot, which is moving around in a workshop. The nine-axis sensor will monitor the attitude angle of the lidar device in real time. For example, it will find that the lidar device has rotated 15° clockwise in the xy plane relative to its initial position (attitude angle change), and at the same time has translated 20 cm upward in the z-axis direction (translation amount).

[0089] In this embodiment, a rigid body transformation matrix for the 3D LiDAR device is constructed based on the device's attitude angles and translations. The rigid body transformation matrix is ​​used to describe the rigid body transformation of the 3D LiDAR device in space, namely, the combined effects of rotation and translation. Using the attitude angle and translation data detected by the nine-axis sensor, this matrix T = [R|t] can be constructed, where R is the rotation matrix and t is the translation vector. This matrix enables the point cloud data currently collected by the 3D LiDAR device to be converted into a unified coordinate system for accurate comparison and analysis with the initial 3D point cloud model. This is because only within the same coordinate system is data collected at different times meaningful and can be effectively processed. Taking the 3D LiDAR device previously used on the mobile robot as an example, based on the detected data of a 15° rotation and a 20 cm upward translation on the z-axis, a rigid body transformation matrix is ​​constructed using a specific mathematical algorithm (such as a combined operation involving a rotation matrix and a translation vector). When the lidar collects new point cloud data, this matrix is ​​used to transform the coordinates of the new data, which is like finding the exact position of the new data in the overall coordinate system, so that it can be aligned with the initial three-dimensional point cloud model established previously and other subsequent operations to ensure the accuracy and consistency of data processing.

[0090] In order to accurately identify all independent dynamic targets in the current point cloud data, it is further proposed to align the current point cloud data with the initial 3D point cloud model and identify all independent dynamic targets, including:

[0091] Use the ICP (Iterative Closest Point) or G-ICP (Generalized Iterative Closest Point) algorithm to align the current point cloud data with the initial 3D point cloud model to obtain the point cloud registration result;

[0092] Separate all dynamic target points in the current point cloud data based on the point cloud registration results;

[0093] Perform cluster analysis on all dynamic target points and identify all independent dynamic targets.

[0094] In this embodiment, the current point cloud data is registered with the initial three-dimensional point cloud model to obtain a point cloud registration result: registration is to find a spatial transformation relationship so that the current point cloud data and the initial three-dimensional point cloud model can be accurately aligned in spatial position. Through registration, the difference between the current scene and the initial modeling can be determined. Suppose in a factory workshop scene, the initial three-dimensional point cloud model depicts the location of fixed facilities such as equipment and passages in the workshop. After a period of time, the three-dimensional lidar device collects the current point cloud data, which may include new personnel or mobile devices entering the workshop. Through a registration algorithm, such as the iterative closest point (ICP) algorithm, the current point cloud data is matched with the initial three-dimensional point cloud model, so that the fixed facility part of the workshop in the current point cloud data is accurately aligned with the corresponding part in the initial model, and finally the point cloud registration result is obtained. This result records the transformation information of the current point cloud data relative to the initial three-dimensional point cloud model.

[0095] In this embodiment, all dynamic target points are separated from the current point cloud data based on the point cloud registration results. By comparing the two sets of registered point cloud data, based on the differences in point cloud characteristics such as position and distribution, point clouds belonging to dynamically changing objects can be identified and extracted. These points are dynamic target points. Taking the factory workshop as an example, after registration, it is found that a portion of the point clouds in the current point cloud data do not overlap with the fixed facility point clouds in the initial three-dimensional point cloud model, and these point clouds exhibit continuous movement characteristics. By further analyzing the spatial distribution and movement trends of these point clouds, it is determined that they belong to forklifts moving within the workshop. Therefore, these point clouds belonging to the forklifts are separated from the current point cloud data, and these points are dynamic target points.

[0096] The system of the present invention supports multiple deployment modes such as fixed installation, vehicle-mounted platform installation, and robot integration, and is applicable to industrial scenarios including but not limited to the following:

[0097] Around dangerous sources such as high-temperature melting furnaces and high-pressure vessels;

[0098] Personnel proximity monitoring in industrial robot operation areas;

[0099] Intrusion detection near high-voltage equipment and lines in power substations;

[0100] Monitoring of AGV logistics channels and personnel interaction areas;

[0101] Warning for personnel accidentally entering the medical radiotherapy room.

[0102] Furthermore, the system boasts comprehensive data collection and storage capabilities. All test results, alarm logs, and device status can be uploaded to the cloud or local servers for easy post-analysis and system optimization. It also supports remote configuration updates and parameter adjustments, reducing the frequency and cost of on-site maintenance.

[0103] like Figure 2 As shown in the figure, in order to fully consider the density and motion characteristics of target points when identifying all independent dynamic targets and adapt to the changes of dynamic targets in different scenarios, it is further proposed to perform cluster analysis on all dynamic target points to identify all independent dynamic targets, including:

[0104] Determine a neighborhood point cloud of each dynamic target point in the current point cloud data, and determine a local point cloud density of each dynamic target point based on the neighborhood point cloud of each dynamic target point;

[0105] An adaptive density threshold is calculated based on the mean and standard deviation of the local point cloud density of all dynamic target points and a preset standard deviation coefficient;

[0106] All dynamic target points whose local point cloud density is not less than the adaptive density threshold are regarded as all core target points;

[0107] Determine the spatial motion joint distance metric data based on the current point cloud and the point cloud at the previous acquisition moment for each core target point;

[0108] Based on the spatial motion joint distance metric data and the local point cloud density of all core target points, multiple core target points are determined as all cluster centers;

[0109] All dynamic target points except all cluster centers are assigned to the cluster center whose local point cloud density is greater than the local point cloud density of the corresponding dynamic target point and is closest to the cluster center, obtaining multiple point clusters, and treating each point cluster as a single independent dynamic target.

[0110] In this embodiment, the neighborhood point cloud of each dynamic target point is determined in the current point cloud data: in the current point cloud data, for each dynamic target point, other point clouds within a certain spatial range surrounding it are determined, and these point clouds constitute the neighborhood point cloud of the dynamic target point. This spatial range can be determined by setting a distance threshold. For example, in a three-dimensional point cloud data space, there is a dynamic target point representing a moving vehicle. With this point as the center, a spherical spatial range with a radius of r is set. All other point clouds within this spherical range are the neighborhood point clouds of the dynamic target point.

[0111] In this embodiment, the local point cloud density of each dynamic target point is determined based on the neighborhood point cloud of each dynamic target point: the local point cloud density is used to measure the density of the point cloud around each dynamic target point. By calculating the number of points contained in the neighborhood point cloud and dividing it by the size of the neighborhood space (such as the volume determined by the previously set spherical neighborhood radius), the local point cloud density of the dynamic target point can be obtained. A higher local point cloud density usually indicates that the point cloud is more concentrated near the point, which may correspond to a part of the object; a lower density indicates that the point cloud is relatively sparse. Continuing with the dynamic target point of the moving vehicle above as an example, if its neighborhood point cloud contains n points in a spherical space with a radius of r, the local point cloud density of the point is calculated as d by a specific formula (such as the number of points divided by the neighborhood volume).

[0112] In this embodiment, the adaptive density threshold is calculated based on the mean and standard deviation of the local point cloud density of all dynamic target points and a preset standard deviation coefficient:

[0113] Explanation: First, the mean of the local point cloud density of all dynamic target points is calculated. This reflects the average level of density across the entire dynamic target point cloud. The standard deviation measures the dispersion of these local point cloud densities relative to the mean. The preset standard deviation coefficient is a manually set parameter used to adjust the sensitivity of the density threshold. An adaptive density threshold is calculated by combining the mean, standard deviation, and the preset standard deviation coefficient (for example, the mean plus the preset standard deviation coefficient multiplied by the standard deviation). This threshold is not fixed but automatically adjusts based on the distribution characteristics of all dynamic target point clouds. This allows it to better adapt to the changing dynamic targets in different scenarios and is used to distinguish core target points from other points in cluster analysis. Assume that the mean of the local point cloud density of all dynamic target points is μ, the standard deviation is σ, and the preset standard deviation coefficient is k. The adaptive density threshold T is calculated by μ + kσ. If the dispersion of the dynamic target point cloud density in a given scenario is large (i.e., the standard deviation σ is large), the calculated adaptive density threshold T will be adjusted accordingly to accommodate this distribution characteristic.

[0114] In this embodiment, all the remaining dynamic target points except all the cluster centers are assigned to the cluster center whose local point cloud density is greater than the local point cloud density of the corresponding dynamic target point and is closest to it, so as to obtain multiple point clusters, and each point cluster is regarded as a single independent dynamic target: after the cluster center has been determined (screened out based on the spatial motion joint distance measurement data and the adaptive density threshold in the previous step), the remaining dynamic target points are assigned to the cluster centers according to certain rules. The specific rule is that each remaining dynamic target point will be assigned to the cluster center that is closest to the dynamic target point among all the cluster centers whose local point cloud density of the cluster center is greater than the local point cloud density of the dynamic target point itself. Through this allocation method, all dynamic target points are divided into multiple point clusters, each point cluster represents a relatively independent dynamic object or target, and the recognition of independent dynamic targets is completed.

[0115] For example, suppose there are three cluster centers C1, C2, and C3, as well as some remaining dynamic target points P1, P2, and P3. For P1, if the local point cloud density of C2 is greater than the local point cloud density of P1, and C2 is the cluster center closest to P1, then P1 is assigned to the point cluster where C2 is located. Similarly, after all remaining dynamic target points are assigned, multiple point clusters are formed. For example, C1 and its assigned points constitute a point cluster, representing an independent dynamic target, perhaps a moving forklift; C2 and its assigned points constitute another point cluster, perhaps representing a walking worker, thereby achieving the division of different independent dynamic targets.

[0116] like Figure 3 As shown in the figure, in order to consider the spatial position relationship and motion difference between core target points when identifying independent dynamic targets in cluster analysis, it is further proposed to determine the spatial motion joint distance measurement data based on the current point cloud and the point cloud of the previous acquisition moment for each core target point, including:

[0117] Determine the three-dimensional coordinates of each core target point in the current point cloud and the point cloud acquired at the previous acquisition moment;

[0118] The vectors pointing from the origin of the preset three-dimensional coordinate system to the three-dimensional coordinates of each core target point in the current point cloud and the point cloud obtained at the previous acquisition moment are regarded as the three-dimensional position vectors of each core target point in the current point cloud and the point cloud obtained at the previous acquisition moment;

[0119] The 2-norm of the difference between the three-dimensional position vectors of each two core target points in the current point cloud is used as the spatial metric value corresponding to the two core target points;

[0120] The difference between the three-dimensional position vectors of each core target point in the current point cloud and the point cloud acquired at the previous acquisition moment is regarded as the inter-frame velocity vector of each core target point;

[0121] The 2-norm of the difference between the velocity vectors of each two core target points is used as the motion distance measurement value corresponding to the two core target points;

[0122] The spatial motion joint distance measurement data includes the spatial measurement values ​​and motion distance measurement values ​​of all pairwise core target points.

[0123] In order to accurately screen out cluster centers with high local point cloud density and large spatial motion joint distance between cluster centers to improve the rationality of clustering results, it is further proposed to determine multiple core target points as all cluster centers based on spatial motion joint distance measurement data and local point cloud density of all core target points, including:

[0124] Determine the defined distance between each two core target points based on the spatial metric value and the motion distance metric value of each two core target points in the spatial motion joint distance metric data;

[0125] Among all core target points, all core target points with a local point cloud density greater than that of a single core target point are selected as a set of high-density candidate target points corresponding to the single core target point.

[0126] The minimum value of the defined distances between each core target point and the corresponding core target point in the large-density candidate target point set of each core target point is taken as the decision distance of the corresponding core target point;

[0127] Based on the decision distance of all core target points, multiple core target points are screened out from all core target points as all cluster centers.

[0128] In this embodiment, the definition distance between each two core target points is determined based on the spatial metric value and motion distance metric value of each two core target points in the spatial motion joint distance metric data: the spatial motion joint distance metric data contains the spatial metric value (reflecting the spatial position difference between them in the current point cloud) and the motion distance metric value (reflecting the difference in their motion state) of each two core target points. In order to comprehensively measure the comprehensive distance between two core target points, it is necessary to integrate these two metric values ​​to determine a distance that can comprehensively reflect the degree of difference between them, namely the definition distance. This definition distance that comprehensively considers the spatial position and motion state can more accurately describe the relationship between the core target points and facilitate the subsequent screening of cluster centers. Assuming that there are core target points A and B, the spatial metric value represents the size of their position difference in the current point cloud. For example, the actual positions of A and B in three-dimensional space are separated by a certain distance. The spatial metric value ds is obtained by calculating the 2-norm of the difference between their three-dimensional position vectors. The motion distance metric value reflects the difference in their motion state. For example, their speed changes at two adjacent acquisition moments are different. The motion distance metric value dm is obtained by calculating the 2-norm of the difference between the velocity vectors between frames. These two values ​​are then combined using some mathematical method (e.g., weighted summation, assuming the weights are ws and wm, respectively, and ws + wm = 1) to obtain the defined distance d = ws × ds + wm × dm. This defined distance d thus comprehensively reflects the difference in spatial position and motion between A and B.

[0129] In this embodiment, multiple core target points are selected from all core target points as all cluster centers based on the decision distances of all core target points: each core target point has a decision distance. A true cluster center should have both high local density and a large decision distance. If the density of a point is very high and there are no points with higher density around it, that is, the decision distance is large, then it is very likely to be a cluster center, but since edge points may also have a large decision distance, in order to avoid filtering edge point clouds when filtering cluster centers, a correction coefficient γ (taken as 0.1) can be introduced to the global maximum value δ′ of the decision distance of all core target points calculated for the first time. m The product term is used to form a recursive relationship, such as calculating the global maximum δ′ of the decision distance of all core target points for the first time. m , and the global maximum value of the decision distance δ′ m , the global maximum value δ of the decision distance between the correction coefficient and all core target points calculated this time m The sum of the product terms of is taken as the revised (second calculated) decision distance δ of the core target point corresponding to the global maximum of the decision distance m (i.e. δ m =δ′ m +γδ m, obtained by transposing ), and then the correction coefficient and the corrected decision distance δ of the core target point corresponding to the global maximum of the decision distance m The product term is introduced into the calculation of the remaining core target points to obtain the second calculated decision distance of all core target points.

[0130] By comparing the second-calculated decision distances of all core target points, we select several core target points with relatively large second-calculated decision distances and reasonable distribution as cluster centers. This can prevent edge points from being selected as cluster centers.

[0131] These cluster centers will be used to cluster all dynamic target points to identify different independent dynamic targets, as they can better represent the characteristics of different dynamic targets. Suppose there are 10 core target points C1, C2, ..., C10, each with its own second-calculated decision distance d1, d2, ..., d10. First, these decision distances are sorted, and then a screening rule is set based on the actual situation. For example, the top three core target points C3, C6, and C8 with the largest second-calculated decision distances are selected as cluster centers. These three cluster centers represent different dynamic target characteristics, and all dynamic target points will be clustered based on them to identify different independent dynamic targets. For example, C3 may represent a person moving towards a dangerous area, and C6 may represent equipment running along a specific track.

[0132] In order to effectively calculate the future motion trend of dangerous dynamic targets, a process for obtaining the motion prediction trajectory of each independent dynamic target is further proposed, including:

[0133] A point cloud data set sequence of each independent dynamic target within the historical period is intercepted from the point cloud data of the complete field environment acquired within the historical period, and a historical motion trajectory of each independent dynamic target is fitted based on the point cloud data set sequence of each independent dynamic target within the historical period;

[0134] A Kalman filter is introduced, and the motion trajectory of each independent dynamic target is predicted based on the historical motion trajectory of each independent dynamic target to determine the motion prediction trajectory of each independent dynamic target.

[0135] In this embodiment, the historical period refers to a set period of time in the past, for example, 10 minutes.

[0136] In this embodiment, the point cloud data set sequence for each independent dynamic target within a historical period is as follows: During the historical period, the point cloud data collected by the 3D LiDAR device contains information about the entire scene. The point cloud data corresponding to each independent dynamic target is extracted and arranged in chronological order, forming a point cloud data set sequence for each independent dynamic target within the historical period. These sequences reflect the spatial position state of the independent dynamic target at different moments within the historical period. Taking the forklift in a logistics warehouse as an example, within a set 10-minute historical period, the LiDAR device collects point cloud data every 10 seconds. From each collected data set, the point cloud data belonging to the forklift is identified and arranged in chronological order according to the time of collection, resulting in a point cloud data set sequence for the forklift within those 10 minutes. For example, the point cloud data set for the forklift collected in the first 10 seconds is S1, the point cloud data set collected in the second 10 seconds is S2, and so on, forming the sequence {S1, S2, ..., S60}.

[0137] In this embodiment, the historical motion trajectory of each independent dynamic target is fitted based on the point cloud data set sequence of each independent dynamic target in the historical period: the point cloud data set sequence of each independent dynamic target in the historical period is processed using mathematical methods to find a curve or path to approximate the motion trajectory of the dynamic target during this period. Linear fitting, polynomial fitting, etc. can be used. For the above-mentioned point cloud data set sequence of the forklift, assuming that the movement of the forklift is roughly a straight line trend, a linear fitting method can be used. By analyzing the coordinate changes of the forklift position points in the point cloud data set sequence in three-dimensional space, a straight line that best fits the distribution trend of these points is found. This straight line is the historical motion trajectory of the forklift. For example, the linear relationship between the position of the forklift in the x, y, and z directions over time is obtained by calculation, thereby determining its historical motion trajectory equation.

[0138] In this embodiment, a Kalman filter is introduced to predict the motion trajectory of each independent dynamic target based on its historical motion trajectory, thereby determining a predicted motion trajectory for each independent dynamic target. By incorporating the Kalman filter into the prediction of the motion trajectory of an independent dynamic target, it continuously updates the prediction of the dynamic target's future position by combining historical motion trajectory data with factors such as system noise and measurement noise. In this way, the next direction and position of the independent dynamic target can be more accurately predicted, allowing for early detection of potential dangerous approaches. Continuing with the example of a forklift, the forklift's historical motion trajectory is known and used as input information for the Kalman filter. The Kalman filter takes into account various interferences that may occur during the forklift's movement (such as changes in ground friction and operator uncertainty, which can be considered noise) to predict the forklift's future motion trajectory. For example, it predicts that the forklift will move to a specific location in a certain direction within the next minute, forming a predicted motion trajectory. This allows for early determination of whether the forklift will approach a dangerous area in the warehouse (such as an area where cargo is stacked too high and could collapse).

[0139] In order to reasonably screen out some core target points used as cluster centers, it is further proposed to determine the dynamic safety distance threshold of each dangerous dynamic target based on the dangerous approach trend and response delay of each independent dynamic target and the additional safety distance of the corresponding danger source area, including:

[0140] Based on the dangerous approach trends of all independent dynamic targets, all dangerous dynamic targets are screened out from all independent dynamic targets;

[0141] Determine the dangerous approach speed of each dangerous dynamic target based on the predicted motion trajectory of each dangerous dynamic target;

[0142] The sum of the response time of the safety device and the response time of the control system and the system delay, the product of the dangerous approach speed of each dangerous dynamic target and the additional safety distance of the corresponding hazard source area is taken as the dynamic safety distance threshold S of the corresponding dangerous dynamic target (i.e., dynamic safety distance threshold S = (sum of response time × approach speed) + additional safety distance). At the same time, factors such as point cloud density and sensor error can also be considered, and d error Dynamically correct safety threshold: d threshold =S+δd error , where d threshold is the modified dynamic safety distance threshold, δ is the correction coefficient, for example, it can be set to 0.25, d error It is the safety threshold effect introduced after considering factors such as point cloud density and sensor error.

[0143] In this embodiment, based on the dangerous approach trends of all independent dynamic targets, all dangerous dynamic targets are screened out from all independent dynamic targets: at the work site, by analyzing the movement of each independent dynamic target, it is determined whether it has a tendency to approach the dangerous source area, which is the dangerous approach trend. Based on this trend, those targets that are more likely to approach the dangerous source area and thus may cause dangerous situations are selected from the many independent dynamic targets. These targets are defined as dangerous dynamic targets. For example, in a factory, independent dynamic targets include workers, transport vehicles, moving robotic arms, etc. Through analysis, it is found that the movement direction of a transport vehicle is gradually approaching the high-temperature furnace area (dangerous source area), and according to its current movement trajectory, it is likely to enter the dangerous range in a short time, so this transport vehicle is screened as a dangerous dynamic target; and the movement direction of another worker is far away from any dangerous source area, so it is not classified as a dangerous dynamic target.

[0144] In this embodiment, the dangerous approach speed of each dangerous dynamic target is determined based on the predicted motion trajectory of each dangerous dynamic target: for an identified dangerous dynamic target, its predicted motion trajectory depicts its possible future motion path. By analyzing this predicted trajectory and combining it with the time factor, the distance that the dangerous dynamic target moves toward the danger source area per unit time is calculated. This distance value is the dangerous approach speed. Assume that the previously identified transport vehicle approaching the high-temperature furnace area, its predicted motion trajectory shows that within the next 5 minutes, it will move from its current position to a place 10 meters away from the furnace. Then, its dangerous approach speed is 10÷5=2 meters / minute.

[0145] By applying Kalman filtering to the target's position in multiple consecutive point cloud frames, the system estimates its trajectory and approach trend. Based on this, the system dynamically sets alarm thresholds and automatically adjusts the warning level based on the target's approach speed, combining the ISO standard's safety distance formula. This method not only improves the system's response speed but also enhances its ability to prevent sudden and dangerous events.

[0146] In order to provide a multi-level intrusion warning mechanism, a corresponding alarm mechanism is further proposed based on the dynamic safety distance threshold of each dangerous dynamic target and the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set, including:

[0147] When the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set is less than 1 times the dynamic safety distance threshold of the corresponding dangerous dynamic target, an audible and visual alarm is triggered;

[0148] When the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set is less than 0.7 times the dynamic safety distance threshold of the corresponding dangerous dynamic target, the linkage control is triggered to limit the equipment action in the corresponding dangerous source area;

[0149] When the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set is less than 0.5 times the dynamic safety distance threshold of the corresponding dangerous dynamic target, the equipment in the corresponding dangerous source area is triggered to shut down urgently and protective measures are started.

[0150] In this embodiment, protective measures refer to a series of targeted actions taken to further ensure personnel safety and reduce potential losses after the equipment emergency shutdown is triggered when the distance between an independent dynamic target and a dangerous point cloud set is less than a specific multiple of the dynamic safety distance threshold of the dangerous dynamic target. For example, in a factory scenario, if a mobile device is detected to be too close to high-voltage electrical equipment (a dangerous source area), triggering the equipment emergency shutdown, the protective measures that may be taken include:

[0151] Evacuate personnel: Initiate emergency evacuation procedures and notify nearby personnel through broadcasts, alarms, etc. to evacuate to a safe area along the predetermined evacuation route to prevent harm to personnel from secondary hazards that may be caused by emergency shutdown of equipment, such as flying equipment parts, leakage, etc.

[0152] Environmental monitoring: For areas where there may be hazardous sources such as toxic and harmful gas leakage and fire risks, activate the corresponding environmental monitoring equipment to monitor the concentration of harmful gases, temperature, smoke and other parameters in the air in real time, so as to promptly detect potential safety hazards and take further measures.

[0153] On-site blockade: Arrange security personnel to block the dangerous area, prevent irrelevant personnel and equipment from entering, and ensure that only professionally trained emergency response personnel can enter the scene to carry out subsequent operations to prevent the accident from escalating.

[0154] A three-level alarm mechanism (sound and light alarm reminder, control linkage limit action, and emergency shutdown protection) has been designed, which can automatically trigger different levels of response measures according to the degree of target proximity, greatly improving safety and automation levels.

[0155] In summary, the present invention effectively solves many problems existing in the existing technology by introducing the multimodal fusion technology of three-dimensional lidar and nine-axis sensor, combining advanced point cloud processing algorithms, trajectory prediction mechanisms and multi-level alarm strategies, and realizes intelligent, dynamic, precise and efficient detection of proximity to dangerous sources. It has broad application prospects and significant social and economic benefits.

[0156] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalents, the present invention is intended to include these modifications and variations.

Claims

1. The operation safety distance management method based on laser radar point cloud data is characterized by: include: Generate a collection of all hazardous point clouds based on the initial 3D point cloud model of the complete scene environment; Perform coordinate transformation on the point cloud data currently collected by the 3D laser radar device based on the rigid body transformation matrix to obtain the current point cloud data; Align the current point cloud data with the initial 3D point cloud model and identify all independent dynamic targets; Determine the dangerous approach trend of each independent dynamic target based on the predicted motion trajectory of each independent dynamic target; Determine a dynamic safety distance threshold for each dangerous dynamic target based on the dangerous approach trend and response delay of each independent dynamic target and the additional safety distance of the corresponding dangerous source area; The corresponding alarm mechanism is triggered based on the dynamic safety distance threshold of each dangerous dynamic target and the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set.

2. The operation safety distance management method based on laser radar point cloud data according to claim 1 is characterized in that: Generate a collection of all hazardous point clouds based on an initial 3D point cloud model of the complete site environment, including: Use 3D LiDAR scanning to integrate the on-site environment and generate an initial 3D point cloud model; All hazard source areas are selected in the initial three-dimensional point cloud model, and the point cloud ranges of all hazard source areas are marked to form a set of all hazardous point clouds.

3. The operation safety distance management method based on laser radar point cloud data according to claim 1 is characterized in that: The process of obtaining the rigid body transformation matrix includes: The attitude angle and translation of the three-dimensional laser radar device are detected in real time based on the nine-axis sensor, and the rigid body transformation matrix of the three-dimensional laser radar device is constructed based on the attitude angle and translation of the three-dimensional laser radar device.

4. The operation safety distance management method based on laser radar point cloud data according to claim 1 is characterized in that: Align the current point cloud data with the initial 3D point cloud model and identify all independent dynamic targets, including: Register the current point cloud data with the initial three-dimensional point cloud model to obtain the point cloud registration result; Separate all dynamic target points in the current point cloud data based on the point cloud registration results; Perform cluster analysis on all dynamic target points and identify all independent dynamic targets.

5. The operation safety distance management method based on laser radar point cloud data according to claim 4 is characterized in that: Perform cluster analysis on all dynamic target points to identify all independent dynamic targets, including: Determine a neighborhood point cloud of each dynamic target point in the current point cloud data, and determine a local point cloud density of each dynamic target point based on the neighborhood point cloud of each dynamic target point; An adaptive density threshold is calculated based on the mean and standard deviation of the local point cloud density of all dynamic target points and a preset standard deviation coefficient; All dynamic target points whose local point cloud density is not less than the adaptive density threshold are regarded as all core target points; Determine the spatial motion joint distance metric data based on the current point cloud and the point cloud at the previous acquisition moment for each core target point; Based on the spatial motion joint distance metric data and the local point cloud density of all core target points, multiple core target points are determined as all cluster centers; All dynamic target points except all cluster centers are assigned to the cluster center whose local point cloud density is greater than the local point cloud density of the corresponding dynamic target point and is closest to the cluster center, obtaining multiple point clusters, and treating each point cluster as a single independent dynamic target.

6. The operation safety distance management method based on laser radar point cloud data according to claim 5 is characterized in that: The spatial motion joint distance metric data is determined based on the current point cloud and the point cloud at the previous acquisition moment for each core target point, including: Determine the three-dimensional coordinates of each core target point in the current point cloud and the point cloud acquired at the previous acquisition moment; The vectors pointing from the origin of the preset three-dimensional coordinate system to the three-dimensional coordinates of each core target point in the current point cloud and the point cloud obtained at the previous acquisition moment are regarded as the three-dimensional position vectors of each core target point in the current point cloud and the point cloud obtained at the previous acquisition moment; The 2-norm of the difference between the three-dimensional position vectors of each two core target points in the current point cloud is used as the spatial metric value corresponding to the two core target points; The difference between the three-dimensional position vectors of each core target point in the current point cloud and the point cloud acquired at the previous acquisition moment is regarded as the inter-frame velocity vector of each core target point; The 2-norm of the difference between the velocity vectors of each two core target points is used as the motion distance measurement value corresponding to the two core target points; The spatial motion joint distance measurement data includes the spatial measurement values ​​and motion distance measurement values ​​of all pairwise core target points.

7. The operation safety distance management method based on laser radar point cloud data according to claim 5 is characterized in that: Based on the spatial motion joint distance metric data and the local point cloud density of all core target points, multiple core target points are determined as all cluster centers, including: Determine the defined distance between each two core target points based on the spatial metric value and the motion distance metric value of each two core target points in the spatial motion joint distance metric data; Among all core target points, all core target points with a local point cloud density greater than that of a single core target point are selected as a set of high-density candidate target points corresponding to the single core target point. The minimum value of the defined distances between each core target point and the corresponding core target point in the large-density candidate target point set of each core target point is taken as the decision distance of the corresponding core target point; Based on the decision distance of all core target points, multiple core target points are screened out from all core target points as all cluster centers.

8. The operation safety distance management method based on laser radar point cloud data according to claim 1 is characterized in that: The process of obtaining the motion prediction trajectory of each independent dynamic target includes: A point cloud data set sequence of each independent dynamic target within the historical period is intercepted from the point cloud data of the complete field environment acquired within the historical period, and a historical motion trajectory of each independent dynamic target is fitted based on the point cloud data set sequence of each independent dynamic target within the historical period; A Kalman filter is introduced, and the motion trajectory of each independent dynamic target is predicted based on the historical motion trajectory of each independent dynamic target to determine the motion prediction trajectory of each independent dynamic target.

9. The operation safety distance management method based on laser radar point cloud data according to claim 1 is characterized in that: The dynamic safety distance threshold of each dangerous dynamic target is determined based on the dangerous approach trend and response delay of each independent dynamic target and the additional safety distance of the corresponding hazard source area, including: Based on the dangerous approach trends of all independent dynamic targets, all dangerous dynamic targets are screened out from all independent dynamic targets; Determine the dangerous approach speed of each dangerous dynamic target based on the predicted motion trajectory of each dangerous dynamic target; The sum of the product of the safety device response time, the control system response time and the system delay, the dangerous approach speed of each dangerous dynamic target and the additional safety distance of the corresponding hazard source area is taken as the dynamic safety distance threshold of the corresponding dangerous dynamic target.

10. The operation safety distance management method based on laser radar point cloud data according to claim 1 is characterized in that: The corresponding alarm mechanism is triggered based on the dynamic safety distance threshold of each dangerous dynamic target and the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set, including: When the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set is less than 1 times the dynamic safety distance threshold of the corresponding dangerous dynamic target, an audible and visual alarm is triggered; When the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set is less than 0.7 times the dynamic safety distance threshold of the corresponding dangerous dynamic target, the linkage control is triggered to limit the equipment action in the corresponding dangerous source area; When the distance between the point cloud set of each independent dynamic target and the dangerous point cloud set is less than 0.5 times the dynamic safety distance threshold of the corresponding dangerous dynamic target, the equipment in the corresponding dangerous source area is triggered to shut down urgently and protective measures are started.

Images