Intelligent video analysis-oriented security method and system for coal preparation plant

Abstract

This invention relates to the field of video recognition and understanding, specifically to a safety method and system for coal preparation plants based on intelligent video analysis. It includes: acquiring video data from the coal preparation plant's work area; extracting objects from video frames and assigning them unique identifiers; extracting key points from objects in the tracking data stream and establishing a dynamic skeleton; calculating the support stability index, center of gravity offset moment, and limb extension tendency based on the dynamic skeleton; identifying the object's behavioral state based on changes in the support stability index, center of gravity offset moment, and limb extension tendency, and determining whether it is an illegal operation state; if an illegal operation state is determined, predicting the future position based on the motion vector of the wrist key point in the dynamic skeleton and calculating the dynamic conflict time; and executing graded early warning and equipment shutdown control. This invention can ensure personnel safety while improving the production efficiency and system reliability of the coal preparation plant.

Description

Technical Field

[0001] This invention relates to the field of video recognition and understanding, and more specifically to a security method and system for coal preparation plants oriented towards intelligent video analysis. Background Technology

[0002] As a crucial link in the coal production process, coal preparation plants exhibit typical industrial characteristics of high risk, dynamics, and complexity. Large mechanical equipment such as crushers, scraper conveyors, filter presses, and high-speed belt conveyors are widely distributed within the plant area. With the deepening implementation of the "smart mine" strategy, intelligent security systems based on video surveillance have become an important means of ensuring personnel safety. However, in actual production processes, operators often need to approach operating equipment to perform necessary operations such as inspections and maintenance. How to accurately and dynamically distinguish between "safe approach" and "dangerous intrusion" has become a core technical challenge that urgently needs to be solved in the current industrial security field.

[0003] Currently, security systems for mechanical operation areas in coal preparation plants generally adopt intelligent monitoring solutions that integrate multiple systems, but significant technical bottlenecks still exist. In existing solutions, the setting and switching of safety warning zones mainly rely on equipment status signals from programmable logic controllers (PLCs), executing static rules in a "lookup table" manner. For example, when the system receives a "power on" signal from the PLC, it loads a preset fixed warning zone; when it receives a "power off" signal, it removes the zone. This method completely ignores the continuous changes in posture and behavioral intentions exhibited by operators during human-machine interaction, causing the system to be unable to effectively distinguish between the following two scenarios at the behavioral level: one is safe "nearby passing," where personnel are conducting normal inspections or passages, approaching the equipment but without any intention to operate; the other is dangerous "leaning forward," where personnel lean forward and extend their arms into the dangerous area of ​​the equipment, posing an extremely high risk of collision.

[0004] Lacking the ability to perceive and analyze the dynamic characteristics of personnel postures, existing systems can only make intrusion judgments based on fixed areas, resulting in a persistently high false alarm rate. Frequent false alarms not only severely disrupt normal production order and reduce inspection efficiency, but also cause trust fatigue among operators towards the security system, weakening its intended early warning and protective effectiveness. Summary of the Invention

[0005] This invention provides a security method and system for coal preparation plants based on intelligent video analytics to solve existing problems.

[0006] The security method for coal preparation plants based on intelligent video analytics of the present invention adopts the following technical solution:

[0007] One embodiment of the present invention provides a security method for coal preparation plants based on intelligent video analytics, the method comprising the following steps:

[0008] Acquire video data from the coal preparation plant's operating area and preprocess it, extract objects from the video frames and assign them unique identifiers, and output a tracking data stream with unique identifiers and real-time positions;

[0009] Key points of objects in the tracking data stream are extracted and a dynamic skeleton is built. Based on the dynamic skeleton, the support stability index, center of gravity offset moment and limb extension tendency are calculated.

[0010] Based on changes in the support stability index, center of gravity offset moment, and limb extension tendency, the behavioral state of the object is identified, and it is determined whether it is an illegal operation state.

[0011] If an illegal operation is detected, the future position is predicted based on the motion vectors of the wrist key points in the dynamic skeleton, and the dynamic conflict time is calculated.

[0012] Based on the relationship between dynamic conflict time, system delay time, and equipment inertia time, tiered early warning and equipment shutdown control are implemented.

[0013] Optionally, objects in video frames are extracted and assigned unique identifiers, and a tracking data stream with unique identifiers and real-time positions is output, specifically including:

[0014] A pre-trained deep learning object detection model is used to perform object detection on pre-processed video frames, identify objects in each frame, and generate bounding boxes and category labels for each detected object.

[0015] Based on at least one of the position, size, and appearance features of the object's bounding box in the preceding and following frames, cross-frame matching and association are performed on the same object, and each object is assigned a unique identifier.

[0016] It continuously tracks the movement trajectory of the object, updates its position coordinates in real time, and forms a tracking data stream with a unique identifier and real-time position information.

[0017] Optionally, key points are extracted from objects in the tracking data stream and a dynamic skeleton is built, specifically including:

[0018] Based on the bounding box of the object in the tracking data stream, the region of the object in the image is located, and human keypoint detection is performed in the region, wherein the human keypoint includes the two-dimensional pixel coordinates of at least one keypoint among the head, neck, shoulder, elbow, wrist, hip, knee and ankle.

[0019] Based on the human physiological structure, key points are logically connected according to preset connection rules to form a dynamic skeleton that represents human posture.

[0020] The dynamic skeleton model is continuously output over time to form a dynamic skeleton data stream for each frame.

[0021] Optionally, the support stability index, center of gravity offset moment, and limb extension tendency are calculated based on the dynamic skeleton, specifically including:

[0022] Based on the positions of key points of the two feet in the dynamic skeleton, the distance between the two feet is calculated, and combined with the human body's center of gravity movement speed, the support stability index is calculated.

[0023] The center of gravity offset moment is calculated based on the horizontal offset distance between the neck key points and the hip key points in the dynamic skeleton, the trunk length, and the trunk tilt angle.

[0024] Based on the distance relationships between key points of the shoulder, elbow, and wrist in the dynamic skeleton, as well as the angle between the arm vector and the horizontal direction, the limb extension tendency is calculated.

[0025] Optionally, based on the positions of key points on both feet in the dynamic skeleton, the distance between the feet is calculated, and combined with the body's center of gravity movement speed, a support stability index is calculated, specifically including:

[0026] Obtain the coordinates of the left and right foot key points;

[0027] Calculate the Euclidean distance between the key points of both feet as the foot spacing;

[0028] Obtain the coordinate sequence of the human body's center of gravity in the preset adjacent frames of the current frame, calculate the average moving speed of the human body's center of gravity per unit time, and obtain the human body's center of gravity moving speed. The preset adjacent frames include the current frame and at least one frame before the current frame.

[0029] The product of the time interval between video frames and the speed at which the human body's center of gravity moves is taken as the theoretical displacement of the human body's center of gravity within the time interval.

[0030] The support stability index is obtained by dividing the distance between the feet by the theoretical displacement of the body's center of gravity during the time interval.

[0031] Optionally, based on the horizontal offset distance between the neck keypoint and the hip keypoint in the dynamic skeleton, the torso length, and the torso tilt angle, the center of gravity offset moment is calculated, specifically including:

[0032] Obtain the horizontal coordinates of the key points on the neck and the key points on the hip;

[0033] Calculate the absolute value of the horizontal coordinate difference between the key points of the neck and the key points of the hip to obtain the horizontal offset distance;

[0034] Obtain the Euclidean distance between the neck key point and the hip key point to get the torso length;

[0035] Based on the positional relationship of key points of the torso in the dynamic skeleton, the angle between the center line of the torso and the vertical direction is calculated to obtain the torso tilt angle.

[0036] Divide the horizontal offset distance by the torso length, and then multiply by the sine of the torso tilt angle to obtain the center of gravity offset moment.

[0037] Optionally, based on the distance relationships between key points of the shoulder, elbow, and wrist in the dynamic skeleton and the angle between the arm vector and the horizontal direction, the limb extension tendency is calculated, specifically including:

[0038] Obtain the coordinates of the shoulder key points, elbow key points, and wrist key points;

[0039] Calculate the Euclidean distance between the shoulder key points and the wrist key points, and use it as the shoulder-wrist distance;

[0040] Calculate the sum of the Euclidean distances between the shoulder key point and the elbow key point, and between the elbow key point and the wrist key point, as the total length of the arm zigzag line;

[0041] Calculate the ratio of the shoulder-wrist distance to the total length of the arm's zigzag line to obtain the distance scaling factor;

[0042] Obtain the cosine of the angle between the arm vector and the horizontal direction;

[0043] Multiplying the distance scaling factor by the cosine value yields the limb extension tendency.

[0044] Optionally, based on changes in the support stability index, center of gravity offset moment, and limb extension tendency, the behavioral state of the object is identified, and it is determined whether it is an illegal operation state, specifically including:

[0045] Obtain time-series data of the supporting stability index within a preset time window. When the data characteristics of the supporting stability index in the time-series data change from fluctuation to being consistently higher than its historical average, the object is determined to change from walking state to stationary state.

[0046] Acquire the time series data of the center of gravity offset moment within a preset time window. When the data characteristics of the center of gravity offset moment in the time series data change from being maintained within the first historical reference range to being continuously higher than its first historical limit value, it is determined that the object's torso is continuously tilting forward toward the device side.

[0047] Acquire time-series data of limb extension tendency within a preset time window. When the data characteristics of limb extension tendency in the time-series data change from within the second historical reference range to continuously exceeding its second historical limit value, and its upward trend overlaps with the upward trend of the center of gravity offset moment in time, it is determined that the person's arm has made a horizontal extension movement towards the equipment side.

[0048] If all of the above judgment results are true within the preset time window, then the behavior state of the judgment object is an illegal operation state.

[0049] Optionally, the future position is predicted based on the motion vectors of wrist keypoints in the dynamic skeleton, and the dynamic conflict time is calculated, specifically including:

[0050] Obtain the normalized coordinates of the wrist key points in the current frame as the current position;

[0051] Obtain the coordinate sequence of wrist key points over at least three consecutive frames;

[0052] Calculate the instantaneous velocity vector of the wrist key points in the current frame based on the coordinate sequence;

[0053] Calculate the instantaneous acceleration vector of the wrist key points in the current frame based on the coordinate sequence;

[0054] Based on the current position, instantaneous velocity vector, and instantaneous acceleration vector, a uniformly accelerated motion model is constructed;

[0055] The future positions of key points on the wrist are calculated using a uniformly accelerated motion model after a preset time period.

[0056] Calculate the minimum distance between the future location and the boundary of the pre-defined danger zone;

[0057] Divide the minimum distance by the component of the wrist's velocity along the direction pointing towards the danger zone to obtain the dynamic conflict time.

[0058] This invention proposes a security system for coal preparation plants based on intelligent video analytics, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the security method for coal preparation plants based on intelligent video analytics.

[0059] The beneficial effects of the technical solution of the present invention are:

[0060] In this embodiment of the invention, by extracting and analyzing the refined posture skeleton and dynamic characteristics of the workers in real time, a dynamic safety boundary risk model based on human behavioral intentions is constructed. This model can accurately distinguish between normal inspection and dangerous operation behaviors and achieve forward-looking prediction of potential collision risks. It significantly reduces the high false alarm rate brought about by traditional systems based on fixed areas and static rules, thereby improving the production efficiency and system reliability of the coal preparation plant while ensuring personnel safety. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 A flowchart of a coal preparation plant security method for intelligent video analytics provided in an embodiment of the present invention;

[0063] Figure 2 This is a structural diagram of a coal preparation plant security system for intelligent video analytics, provided as an embodiment of the present invention. Detailed Implementation

[0064] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the coal preparation plant security method for intelligent video analysis proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0065] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0066] The following description, in conjunction with the accompanying drawings, details the specific scheme of the coal preparation plant security method for intelligent video analysis provided by this invention.

[0067] This invention provides a security method and system for coal preparation plants based on intelligent video analytics. Please refer to [link / reference]. Figure 1 The diagram illustrates a flowchart of a security method for coal preparation plants based on intelligent video analytics, according to an embodiment of the present invention. The method includes the following steps:

[0068] S101. Acquire video data from the coal preparation plant's operating area and preprocess it. Extract objects from the video frames and assign them unique identifiers. Output a tracking data stream with unique identifiers and real-time positions.

[0069] In this embodiment, objects are extracted from video frames and assigned unique identifiers, and a tracking data stream with unique identifiers and real-time positions is output, specifically including:

[0070] A pre-trained deep learning object detection model is used to perform object detection on pre-processed video frames, identify objects in each frame, and generate bounding boxes and category labels for each detected object.

[0071] Based on at least one of the position, size, and appearance features of the object's bounding box in the preceding and following frames, cross-frame matching and association are performed on the same object, and each object is assigned a unique identifier.

[0072] It continuously tracks the movement trajectory of the object, updates its position coordinates in real time, and forms a tracking data stream with a unique identifier and real-time position information.

[0073] For example, preprocessing includes, but is not limited to, at least one of the following: dustproofing and light-adaptive enhancement, motion blur compensation, and dynamic noise filtering of video data, in order to improve the usability of video images in the complex lighting and dust environment of a coal preparation plant.

[0074] Furthermore, objects in the video frames are extracted and assigned unique identifiers, and a tracking data stream with unique identifiers and real-time positions is output, specifically including:

[0075] The pre-trained deep learning object detection model (such as YOLOv5, Faster R-CNN or its industrial-optimized version) is used to process the pre-processed video frames, identify objects such as "people", "equipment", and "materials" in each frame, and generate bounding boxes and their corresponding category labels for each detected object.

[0076] Based on at least one metric among the overlap of object bounding boxes between adjacent frames, motion continuity, and appearance feature similarity, multiple detection boxes belonging to the same target are matched and associated, and a globally unique identifier (ID) is assigned and maintained for each continuously tracked object.

[0077] For each identified object, its position coordinates (such as the center point of the bounding box) in each frame are continuously recorded to form a temporal motion trajectory, and output in the form of a structured data stream. Each record includes at least: timestamp, object ID, object category, and current position coordinates.

[0078] Optionally, if an object corresponding to a certain ID is not detected again in several consecutive frames, the system will release the ID after a certain delay and remove it from the tracking list to avoid ID resource exhaustion and misassociation.

[0079] S102. Extract key points of objects in the tracking data stream and build a dynamic skeleton. Calculate the support stability index, center of gravity offset moment, and limb extension tendency based on the dynamic skeleton.

[0080] In this embodiment, key points are extracted from objects in the tracking data stream and a dynamic skeleton is built, specifically including:

[0081] Based on the bounding box of the object in the tracking data stream, the region of the object in the image is located, and human keypoint detection is performed in the region, wherein the human keypoint includes the two-dimensional pixel coordinates of at least one keypoint among the head, neck, shoulder, elbow, wrist, hip, knee and ankle.

[0082] Based on the human physiological structure, key points are logically connected according to preset connection rules to form a dynamic skeleton that represents human posture.

[0083] The dynamic skeleton model is continuously output over time to form a dynamic skeleton data stream for each frame.

[0084] The dynamic skeleton is used to calculate the support stability index, center of gravity offset moment, and limb extension tendency, specifically including:

[0085] Based on the positions of key points of the two feet in the dynamic skeleton, the distance between the two feet is calculated, and combined with the human body's center of gravity movement speed, the support stability index is calculated.

[0086] The center of gravity offset moment is calculated based on the horizontal offset distance between the neck key points and the hip key points in the dynamic skeleton, the trunk length, and the trunk tilt angle.

[0087] Based on the distance relationships between key points of the shoulder, elbow, and wrist in the dynamic skeleton, as well as the angle between the arm vector and the horizontal direction, the limb extension tendency is calculated.

[0088] Based on the positions of key points on both feet in the dynamic skeleton, the distance between the feet is calculated, and combined with the body's center of gravity movement speed, the support stability index is calculated, specifically including:

[0089] Obtain the coordinates of the left and right foot key points;

[0090] Calculate the Euclidean distance between the key points of both feet as the foot spacing;

[0091] Obtain the coordinate sequence of the human body's center of gravity in the preset adjacent frames of the current frame, calculate the average moving speed of the human body's center of gravity per unit time, and obtain the human body's center of gravity moving speed. The preset adjacent frames include the current frame and at least one frame before the current frame.

[0092] The product of the time interval between video frames and the speed at which the human body's center of gravity moves is taken as the theoretical displacement of the human body's center of gravity within the time interval.

[0093] The support stability index is obtained by dividing the distance between the feet by the theoretical displacement of the body's center of gravity during the time interval.

[0094] Based on the horizontal offset distance between the neck keypoint and the hip keypoint in the dynamic skeleton, the torso length, and the torso tilt angle, the center of gravity offset moment is calculated, specifically including:

[0095] Obtain the horizontal coordinates of the key points on the neck and the key points on the hip;

[0096] Calculate the absolute value of the horizontal coordinate difference between the key points of the neck and the key points of the hip to obtain the horizontal offset distance;

[0097] Obtain the Euclidean distance between the neck key point and the hip key point to get the torso length;

[0098] Based on the positional relationship of key points of the torso in the dynamic skeleton, the angle between the center line of the torso and the vertical direction is calculated to obtain the torso tilt angle.

[0099] Divide the horizontal offset distance by the torso length, and then multiply by the sine of the torso tilt angle to obtain the center of gravity offset moment.

[0100] Based on the distance relationships between key points at the shoulder, elbow, and wrist in the dynamic skeleton, and the angle between the arm vector and the horizontal direction, the limb extension tendency is calculated, specifically including:

[0101] Obtain the coordinates of the shoulder key points, elbow key points, and wrist key points;

[0102] Calculate the Euclidean distance between the shoulder key points and the wrist key points, and use it as the shoulder-wrist distance;

[0103] Calculate the sum of the Euclidean distances between the shoulder key point and the elbow key point, and between the elbow key point and the wrist key point, as the total length of the arm zigzag line;

[0104] Calculate the ratio of the shoulder-wrist distance to the total length of the arm's zigzag line to obtain the distance scaling factor;

[0105] Obtain the cosine of the angle between the arm vector and the horizontal direction;

[0106] Multiplying the distance scaling factor by the cosine value yields the limb extension tendency.

[0107] For example, based on a tracking data stream, human keypoint detection and skeleton construction are performed on objects categorized as "human," the process including:

[0108] Based on the obtained personnel bounding box, a local area image centered on the box is cropped from the original image to eliminate interference from complex dynamic backgrounds such as drum rotation and coal flow, and to concentrate computational resources on human posture analysis.

[0109] On local image regions, deep learning-based keypoint detection models (such as OpenPose, HRNet, AlphaPose, or improved models suitable for industrial workwear scenarios) are used to detect multiple representative anatomical feature points of the human body, including:

[0110] Core points of the torso: center of the head, neck, left shoulder, right shoulder, left hip, right hip;

[0111] Limb control points: left elbow, right elbow, left wrist, right wrist, left knee, right knee, left ankle, right ankle.

[0112] The model outputs the precise pixel coordinates of each of the above key points in the image coordinate system.

[0113] The detected key point coordinates are logically connected according to the natural physiological structure of the human body to form a structured human skeleton representation. The connections include, but are not limited to: the neck connecting to the left and right shoulders, the left shoulder connecting to the left elbow, the left elbow connecting to the left wrist, the right shoulder connecting to the right elbow, the right elbow connecting to the right wrist, the left hip connecting to the left knee, the left knee connecting to the left ankle, the right hip connecting to the right knee, and the right knee connecting to the right ankle, etc.

[0114] The key point coordinates and their connections for each tracked person in each frame are combined to form the dynamic skeleton data of that person in that frame, and then output as a time-series data stream.

[0115] Optionally, to address the issue of blurred outlines that may be caused by workers wearing heavy work clothes in coal preparation plants, the key point detection model can be fine-tuned using a training dataset that enhances the outline of work clothes, thereby improving the robustness of joint localization.

[0116] After acquiring dynamic skeleton data, the system analyzes the geometric and motion relationships between key points in the skeleton to quantitatively identify personnel behavior, distinguishing between "normal inspection" and "violations and risky behaviors." Specifically, this includes:

[0117] Lower limb movement pattern analysis:

[0118] Gait can be determined by the temporal displacement changes of key points on both feet (such as the ankle).

[0119] Normal inspection characteristics: The key points of the left and right feet show periodic alternation, indicating that the person is in a walking state.

[0120] Disqualification signs: The movement of the two feet at key points stops alternating and maintains a relatively fixed, large distance (resembling a "stepping" standing posture), indicating that the person has stopped moving and entered a static support preparation state.

[0121] Trunk posture stability analysis:

[0122] Assess the body's center of gravity shift based on the connection between key points of the spine (such as the head, neck, and hips).

[0123] Normal inspection characteristics: The angle between the line connecting the key points of the spine and the vertical direction is small, and the fluctuation range during walking is limited, indicating that the body's center of gravity is stable and there is no significant tendency to lean forward.

[0124] Disqualifying signs: A continuous and significant deflection of the line connecting key points of the spine toward the equipment side (such as the direction of a belt conveyor or roller) indicates that the body's center of gravity has shifted forward, creating a tendency to lean forward.

[0125] Analysis of upper limb movement intentions:

[0126] Based on the geometric relationship and direction of movement of key arm points (shoulder, elbow, wrist), the intention of arm behavior can be determined.

[0127] Normal inspection characteristics: The key points of the wrist are mainly on the sides of the body or in front of the chest, and its movement trajectory is within the safe range, maintaining a certain distance from the dangerous areas of the equipment.

[0128] Characteristics of the violation:

[0129] Geometric features: The shoulder, elbow, and wrist are close to a straight line, indicating that the arm is fully extended;

[0130] Directional characteristics: The direction vector of the wrist's key movement points is horizontal and points into the dangerous area of ​​the equipment (such as belt gaps and roller gaps).

[0131] Unlike compliant actions: when bending over to pick up an item, the wrist's key point moves mainly vertically downwards; while when illegally reaching out, the wrist's key point moves horizontally inwards, forming a "plunging" trajectory.

[0132] If the system detects that the following three characteristics occur consecutively in time: the lower limbs enter a static support preparation state, the torso leans significantly forward toward the equipment, and the upper limbs extend horizontally and point toward the dangerous area of ​​the equipment, then the current personnel behavior state is determined to be a violation and risk (illegal operation state), and subsequent risk prediction and early warning processes are triggered.

[0133] Optionally, the system can adaptively learn behavioral benchmarks such as gait and posture of different personnel based on historical inspection data, reducing misjudgments due to individual differences and improving the personalization and accuracy of status recognition.

[0134] For example, in order to eliminate scale differences caused by varying distances between people and cameras, and to achieve robust behavior quantification across scenes, this embodiment first normalizes the skeleton coordinate system before performing specific index calculations. The specific steps are as follows:

[0135] The origin of the coordinate system is the center point of the hip of the person in the current frame; the unit length is the length of the person's torso (the Euclidean distance from the key point of the neck to the center point of the hip); the horizontal direction is set as the X-axis and the vertical direction is set as the Y-axis to form a scale-invariant human body relative coordinate system; the coordinates of all key points of the human body (such as shoulders, elbows, wrists, knees, ankles, etc.) are transformed to this relative coordinate system.

[0136] Specifically, the stability index The calculation formula can be:

[0137]

[0138] in, and These are the coordinates of the left and right ankles in a relative coordinate system, respectively. Indicates the time interval between adjacent frames in a video; It represents the average movement speed of the human body's center of gravity (center point of the hip) over the most recent consecutive frames (such as the first 5 frames).

[0139] Alternatively, the process of obtaining the velocity of the human body's center of gravity in the above formula can be as follows:

[0140] Record the pixel coordinates of the human body's center of gravity in the current frame and the previous K consecutive frames (K≥2) to form a coordinate sequence, where the human body's center of gravity is represented by the coordinates of the hip center point in the dynamic skeleton;

[0141] Calculate the Euclidean distance between the center points of the human body in adjacent frames;

[0142] Sum the above K-1 displacements and divide by the corresponding time span to obtain the average speed of the human body's center of gravity moving per unit time.

[0143] The calculated average movement speed is output as the human body's center of gravity movement speed in the current frame, which is used to support the calculation of the stability index SSI.

[0144] The calculation logic is as follows: When a person walks, the distance between their feet changes periodically, the center of gravity velocity is not zero, and the SSI value fluctuates dramatically; when a person is about to perform an unauthorized operation, the distance between their feet stabilizes at a larger value, and the center of gravity velocity approaches zero, the SSI value stabilizes at a higher level, indicating that they have entered a static support state. This index characterizes the support stability provided by the feet per unit displacement velocity; the higher the SSI value, the closer the person is to a static support state.

[0145] Center of gravity offset moment The calculation formula is:

[0146]

[0147] in, and These are the coordinates of the key points of the neck and the center point of the hip on the X-axis, respectively. This refers to the length of the torso (per unit length). It is the angle between the center line of the torso (from the neck to the hip) and the vertical direction (the negative direction of the Y-axis).

[0148] The higher the CGO value, the more pronounced the forward tilt of the torso towards the equipment side, and the stronger the tendency to lean forward.

[0149] Limb extension tendency The calculation formula is:

[0150]

[0151] in, , and These are the shoulder-wrist distance, shoulder-elbow distance, and elbow-wrist distance, respectively. The angle between the vector of the line connecting the shoulder to the wrist and the horizontal direction (X-axis).

[0152] When the arm is fully extended and horizontally stretched, the numerator approaches the denominator. The LEI value approaches 1; when the arm is bent or pointing vertically downwards, the numerator is smaller than the denominator. The LEI value is relatively small, significantly less than 1; therefore, the higher the LEI value, the stronger the tendency for the arm to extend horizontally and towards the side of the device.

[0153] If the system identifies a person in a statically supported position (high and stable SSI value), with a clearly forward-leaning torso (continuously rising CGO value) and arms extended horizontally (LEI value approaching 1), it will be comprehensively judged as a high-risk "leaning operation" behavior.

[0154] Optionally, the threshold values ​​of each indicator can be adaptively calibrated according to the specific work scenario, equipment type and personnel posture, so as to further improve the adaptability and accuracy of the system under different working conditions.

[0155] S103. Based on the changes in the support stability index, center of gravity offset moment, and limb extension tendency, identify the object's behavioral state and determine whether it is an illegal operation state.

[0156] In this embodiment, the behavioral state of the object is identified based on changes in the support stability index, center of gravity offset moment, and limb extension tendency, and it is determined whether the operation is illegal. Specifically, this includes:

[0157] Obtain time-series data of the supporting stability index within a preset time window. When the data characteristics of the supporting stability index in the time-series data change from fluctuation to being consistently higher than its historical average, the object is determined to change from walking state to stationary state.

[0158] Acquire the time series data of the center of gravity offset moment within a preset time window. When the data characteristics of the center of gravity offset moment in the time series data change from being maintained within the first historical reference range to being continuously higher than its first historical limit value, it is determined that the object's torso is continuously tilting forward toward the device side.

[0159] Acquire time-series data of limb extension tendency within a preset time window. When the data characteristics of limb extension tendency in the time-series data change from within the second historical reference range to continuously exceeding its second historical limit value, and its upward trend overlaps with the upward trend of the center of gravity offset moment in time, it is determined that the person's arm has made a horizontal extension movement towards the equipment side.

[0160] If all of the above judgment results are true within the preset time window, then the behavior state of the judgment object is an illegal operation state.

[0161] For example, to achieve personalized behavior recognition for different individuals and improve system adaptability, this embodiment introduces a personal behavior benchmark mechanism, specifically including:

[0162] A behavioral baseline window is maintained for each tracked individual in real time, continuously recording the SSI, CGO, and LEI index values ​​for the past N consecutive frames (e.g., 50 frames, corresponding to approximately 2 seconds); the moving average of each index within the time window is calculated as the individual's personal behavioral baseline value for the current stage.

[0163] Based on the comparison between current indicator values ​​and individual behavioral benchmarks, personnel behaviors are categorized as follows:

[0164] Standard inspection mode:

[0165] Judgment criteria: SSI fluctuates regularly around its baseline value; CGO remains at a historically low level (below the baseline value + preset threshold); LEI has a very small rate of change (e.g., the change is below the threshold). The system is deemed safe, and individual behavioral baselines are continuously updated to accommodate individuals of different heights, gaits, and work habits.

[0166] Compliant bending over / low-risk movements:

[0167] Judgment criteria: The SSI changes from a fluctuating state to a stable state and is significantly higher than the benchmark value, indicating that both feet have entered static support; the CGO shows a significant positive deviation (consistently higher than the benchmark value + preset threshold), indicating forward trunk tilt; however, the LEI does not rise synchronously and remains at a low level, and Small (arm vector nearly vertical). Determined as a compliant bending motion (such as picking up an item), and does not trigger a danger alarm.

[0168] Illegal leaning-out operation (high-risk situation):

[0169] Judgment criteria: SSI is consistently and significantly higher than the baseline value, indicating that both feet are "locked" to the ground; CGO is consistently rising and significantly higher than the baseline value, accompanied by a directional trend towards the device (determined through vector analysis); LEI and CGO show a strong positive correlation and synchronous growth, that is, as the torso leans forward, the arms extend horizontally synchronously, and Approaching 1. When the above three characteristics deviate significantly simultaneously within the same time window (e.g., 0.5 seconds), the system determines it to be in an "illegal body-leaning operation" state and immediately activates the subsequent risk prediction and early warning mechanism.

[0170] During the period when personnel are identified as being in "routine inspection mode", the system continuously updates their behavioral benchmarks using a sliding window to ensure that the benchmark values ​​can follow the natural changes in the personnel's work status (such as changes in walking speed, short stops, etc.), thereby improving the system's adaptability to different work stages and personnel.

[0171] Optionally, different first historical benchmark ranges, first historical upper limits, second historical benchmark ranges, second historical upper limits, preset thresholds, etc., can be set and dynamically adjusted according to equipment type and risk level of work area to achieve more refined risk response.

[0172] For example, the stability state of the support can be determined as follows:

[0173] Acquire time-series data supporting the stability index within a preset time window (e.g., the most recent 50 frames), and calculate its statistical characteristics, including but not limited to mean, variance, and range. When the variance of the time-series data is detected to decrease significantly and the value is consistently higher than the historical average value within that time period, reaching a first preset threshold, and the fluctuation range (range) of the data points converges to a low range, it is determined that the object has changed from a walking state to a stationary state.

[0174] Determination of forward-leaning posture:

[0175] Acquire time-series data of the center-of-gravity offset moment within the same time window, and establish its first historical baseline range (e.g., mean ± standard deviation) within that time period. When the time-series data changes from small fluctuations around the baseline to a continuous unidirectional increase, and the value exceeds the first historical limit for several consecutive frames (e.g., more than 5 frames) (e.g., mean + 2 times standard deviation), it is determined that the object's torso is exhibiting a continuous forward tilt towards the device.

[0176] Determining the horizontal extension position of the arm:

[0177] Acquire time-series data on limb extension tendency within the same time window and establish a second historical baseline range for that time period. When the time-series data changes from being stable or fluctuating slightly within the baseline range to a continuous increase, and the values ​​of several consecutive frames exceed the second historical limit, calculate the temporal correlation between this upward trend and the upward trend of the center of gravity offset moment. If the two show a significant positive correlation within the time window (correlation coefficient greater than a set threshold, such as 0.7), and the start time and peak time of the upward process overlap, then it is determined that the person's arm has performed a horizontal extension movement towards the equipment side.

[0178] If all three judgment results are true simultaneously within the same time window (e.g., within 2 seconds), the object is considered to be in an "illegal operation state." Otherwise, it is judged to be in a safe or low-risk state.

[0179] "Volatility" refers to the non-stationary, periodic, or random change patterns of time-series data in a statistical sense, which can be quantified using one or more of the following indicators:

[0180] Significant variance / standard deviation: Within a preset time window, the variance or standard deviation of the time series data is consistently higher than the set fluctuation threshold, indicating that the data is highly dispersed and exhibits an unstable state.

[0181] Periodic variation in range: The difference between the maximum and minimum values ​​of time series data exhibits regular fluctuations. For example, the alternating movement of the feet during walking causes the support stability index to exhibit high-frequency periodic oscillations.

[0182] Weak autocorrelation: The autocorrelation coefficient of time series data is close to zero or negative within a short time delay (such as 1-3 frames), indicating that there is a lack of continuity between data at adjacent time points, reflecting the incoherence of actions;

[0183] Frequency domain energy distribution: Perform Fourier transform or wavelet transform on time series data. If there is a significant energy peak in the step frequency correlation band (such as 1-2 Hz), it represents the periodic fluctuation in the walking state.

[0184] The criteria for determining the fluctuation state can be further refined as follows:

[0185] Time series data is considered to be in a "fluctuation state" when the following conditions are met simultaneously: the variance is greater than N times (e.g., 1.5 times) the historical baseline variance; the range is consistently higher than the threshold and exhibits periodicity; and the autocorrelation coefficient does not show a significant positive correlation within the step frequency period.

[0186] When the fluctuation characteristics disappear, the following are observed: the variance drops rapidly to below the threshold; the data sequence tends to be stable and the range converges; the autocorrelation coefficient becomes continuously positive, reflecting that the state is maintained.

[0187] S104. If the operation is determined to be illegal, predict the future position based on the motion vector of the wrist key point in the dynamic skeleton and calculate the dynamic conflict time.

[0188] In this embodiment, the future position is predicted based on the motion vectors of key wrist points in the dynamic skeleton, and the dynamic conflict time is calculated, specifically including:

[0189] Obtain the normalized coordinates of the wrist key points in the current frame as the current position;

[0190] Obtain the coordinate sequence of wrist key points over at least three consecutive frames;

[0191] Calculate the instantaneous velocity vector of the wrist key points in the current frame based on the coordinate sequence;

[0192] Calculate the instantaneous acceleration vector of the wrist key points in the current frame based on the coordinate sequence;

[0193] Based on the current position, instantaneous velocity vector, and instantaneous acceleration vector, a uniformly accelerated motion model is constructed;

[0194] The future positions of key points on the wrist are calculated using a uniformly accelerated motion model after a preset time period.

[0195] Calculate the minimum distance between the future location and the boundary of the pre-defined danger zone;

[0196] Divide the minimum distance by the component of the wrist's velocity along the direction pointing towards the danger zone to obtain the dynamic conflict time.

[0197] For example, when the system determines that a person is in an "illegal operation state," it immediately activates a risk prediction mechanism. By modeling and analyzing the movement trajectory of key points on their wrist, it proactively assesses potential collision risks. The specific steps are as follows:

[0198] Extract the continuous coordinate sequence of key points on the person's wrist within the time window when the operation was deemed "illegal".

[0199] Based on this sequence, the instantaneous velocity vector (first derivative) and instantaneous acceleration vector (second derivative) of the wrist key points in each frame are calculated to form a set of motion state descriptions with the wrist as the core.

[0200] Normalized coordinates of wrist key points in the current frame Current velocity vector Current acceleration vector Using a uniformly accelerated motion model to predict a future moment (Typically set to 0.5 seconds) Future position of key points on the wrist Therefore, based on the above parameters, the uniformly accelerated motion model can be constructed as follows:

[0201]

[0202] Collision risk prediction: A 3D model of the corresponding "dangerous no-go zone" for the equipment (such as the belt conveyor roller area, crusher feed inlet, etc.) is preset, and the predicted location is... Mapped to the coordinate system of the model; if If the wrist falls into a predefined "danger zone", it is considered a potential collision, even if the wrist's current actual coordinates have not yet entered that area.

[0203] For example, the current coordinates of the person's right wrist are (200, 300), the velocity is (50px / s, 0), and the acceleration is ( The predicted position after 0.5 seconds is (215, 300).

[0204] If the predicted location falls within the 'danger zone' model of the conveyor belt roller, the system determines that a collision may occur in 0.5 seconds and immediately triggers an early warning.

[0205] Optionally, the system can set different prediction times based on different equipment types and hazard levels. For high-speed equipment, the prediction time can be shortened to improve response timeliness, while for low-speed or high-inertia equipment, the prediction time can be appropriately extended to smooth out misjudgments.

[0206] S105. Based on the relationship between dynamic conflict time, system delay time and equipment inertia time, implement graded early warning and equipment shutdown control.

[0207] In this embodiment, based on the relationship between dynamic conflict time, system delay time, and equipment inertial time, graded early warning and equipment shutdown control are performed, specifically including:

[0208] Add the system delay time to the device inertia time to obtain the total system response time;

[0209] Compare dynamic conflict time with total system response time:

[0210] If the dynamic conflict time exceeds the total system response time, the audible and visual warning device will be triggered to issue a warning.

[0211] If the dynamic conflict time is less than or equal to the total system response time, an emergency stop command is sent to the equipment control system.

[0212] For example, when the system determines that a potential collision has occurred, it further calculates the dynamic conflict time and, based on a comparison with the system's response capability, executes a graded warning or emergency braking. The specific steps are as follows:

[0213] Calculate dynamic conflict time The calculation formula can be:

[0214]

[0215] in, This represents the Euclidean distance from the current wrist point to the nearest dangerous physical boundary (such as the edge of a safety net or the surface of a roller). This indicates the magnitude of the current wrist movement speed; Indicates the angle between the direction of movement of the key points of the wrist and the line pointing towards the center of the danger zone; It represents a very small positive number and is used to prevent the denominator from being zero.

[0216] A lower TTC value indicates a more imminent danger. If the TTC continues to decrease rapidly, it indicates that personnel are accelerating and approaching the hazard in a directed manner.

[0217] The system compares TTC (Total Time Tolerance) with the total system response time to make a tiered decision. (Total System Response Time) It consists of the following two parts ( ):

[0218] System processing latency, including the time from video processing, skeleton recognition, decision calculation to instruction generation (e.g., 0.05 seconds).

[0219] Mechanical inertia time of the equipment, which is the time required from receiving the stop command to complete stop (obtained through actual measurement of the equipment).

[0220] The decision-making rules are as follows:

[0221]

[0222] Execute corresponding controls based on the RIC decision value:

[0223] Tiered early warning (RIC=0, but TTC is declining): The system triggers the audible and visual warning device, broadcasting a voice prompt (e.g., "Danger area, please retreat immediately!") through on-site loudspeakers, while simultaneously flashing a warning light. This aims to prevent danger through proactive personnel response and reduce unnecessary downtime.

[0224] Rigid Stop (RIC=1): The system immediately sends a high-priority emergency stop command to the PLC (Programmable Logic Controller) via Industrial Ethernet. This is because mechanical inertia time has been compensated for in the decision-making process. The equipment can completely stop before personnel actually come into contact with the danger, achieving physical-level safety isolation.

[0225] Each warning or shutdown event is recorded by the system, including information such as time, personnel ID, hazard type, predicted trajectory, TTC value, and actions performed, for post-event analysis and system optimization.

[0226] In summary, in this embodiment of the invention, video data is acquired by industrial cameras deployed in key operational areas of the coal preparation plant. After preprocessing and using a deep learning target detection model, personnel targets are extracted and a tracking data stream with a unique identifier is generated. Furthermore, a dynamic skeleton is constructed based on human keypoint detection technology. By calculating three core behavioral indicators—Support Stability Index (SSI), Center of Gravity Offset Moment (CGO), and Limb Extension Inclination (LEI)—and establishing an individual adaptive behavioral benchmark, accurate identification of dangerous behaviors such as "illegal leaning over" is achieved. Based on this, the future position is predicted using the motion state of wrist keypoints, and the Dynamic Conflict Time (TTC) is calculated. Combined with system latency and equipment braking inertia time, a tiered early warning and emergency shutdown decision-making mechanism is constructed. Finally, through the linkage control of audible and visual warnings and a programmable logic controller (PLC), a closed-loop security system is achieved, encompassing dangerous situation perception, behavioral intent recognition, collision risk prediction, and proactive safety intervention. This significantly improves the safety, early warning accuracy, and system response time of human-machine collaborative operations in the coal preparation plant, effectively solving the technical bottlenecks of high false alarm rates and lack of behavioral prediction capabilities in traditional fixed-area monitoring.

[0227] This invention also proposes a security system for coal preparation plants based on intelligent video analytics; please refer to [link / reference]. Figure 2 The diagram shows a structural diagram of a coal preparation plant security system for intelligent video analysis provided by an embodiment of the present invention. The system includes: a data acquisition module 101, a data processing module 102, and an early warning control module 103.

[0228] The data acquisition module 101 is used to acquire video data from the coal preparation plant's operating area and preprocess it, extract objects from the video frames and assign them unique identifiers, and output a tracking data stream with unique identifiers and real-time positions.

[0229] Data processing module 102 is used to extract key points of objects in the tracking data stream and build a dynamic skeleton, and calculate the support stability index, center of gravity offset moment and limb extension tendency based on the dynamic skeleton.

[0230] Based on changes in the support stability index, center of gravity offset moment, and limb extension tendency, the behavioral state of the object is identified, and it is determined whether it is an illegal operation state.

[0231] If an illegal operation is detected, the future position is predicted based on the motion vectors of the wrist key points in the dynamic skeleton, and the dynamic conflict time is calculated.

[0232] The early warning control module 103 is used to perform graded early warning and equipment shutdown control based on the relationship between dynamic conflict time, system delay time and equipment inertia time.

[0233] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer equipment can be divided into different functional modules to complete all or part of the functions described above. In addition, the coal preparation plant security system for intelligent video analysis and the coal preparation plant security method embodiment for intelligent video analysis provided in the above embodiments belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0234] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0235] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0236] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A security method for coal preparation plants based on intelligent video analytics, characterized in that, include: Acquire video data from the coal preparation plant's operating area and preprocess it, extract objects from the video frames and assign them unique identifiers, and output a tracking data stream with unique identifiers and real-time positions; Key points of objects in the tracking data stream are extracted and a dynamic skeleton is built. Based on the dynamic skeleton, the support stability index, center of gravity offset moment and limb extension tendency are calculated. Based on changes in the support stability index, center of gravity offset moment, and limb extension tendency, the behavioral state of the object is identified, and it is determined whether it is an illegal operation state. If an illegal operation is detected, the future position is predicted based on the motion vectors of the wrist key points in the dynamic skeleton, and the dynamic conflict time is calculated. Based on the relationship between dynamic conflict time, system delay time, and equipment inertia time, tiered early warning and equipment shutdown control are implemented.

2. The coal preparation plant security method based on intelligent video analytics according to claim 1, characterized in that, The process of extracting objects from video frames, assigning them unique identifiers, and outputting a tracking data stream with unique identifiers and real-time positions specifically includes: A pre-trained deep learning object detection model is used to perform object detection on pre-processed video frames, identify objects in each frame, and generate bounding boxes and category labels for each detected object. Based on at least one of the position, size, and appearance features of the object's bounding box in the preceding and following frames, cross-frame matching and association are performed on the same object, and each object is assigned a unique identifier. It continuously tracks the movement trajectory of the object, updates its position coordinates in real time, and forms a tracking data stream with a unique identifier and real-time position information.

3. The coal preparation plant security method based on intelligent video analytics according to claim 1, characterized in that, The process of extracting key points from objects in the tracking data stream and building a dynamic skeleton specifically includes: Based on the bounding box of the object in the tracking data stream, the region of the object in the image is located, and human keypoint detection is performed in the region, wherein the human keypoint includes the two-dimensional pixel coordinates of at least one keypoint among the head, neck, shoulder, elbow, wrist, hip, knee and ankle. Based on the human physiological structure, key points are logically connected according to preset connection rules to form a dynamic skeleton that represents human posture. The dynamic skeleton model is continuously output over time to form a dynamic skeleton data stream for each frame.

4. The coal preparation plant security method based on intelligent video analytics according to claim 1, characterized in that, The calculation of the support stability index, center of gravity offset moment, and limb extension tendency based on the dynamic skeleton specifically includes: Based on the positions of key points of the two feet in the dynamic skeleton, the distance between the two feet is calculated, and combined with the human body's center of gravity movement speed, the support stability index is calculated. The center of gravity offset moment is calculated based on the horizontal offset distance between the neck key points and the hip key points in the dynamic skeleton, the trunk length, and the trunk tilt angle. Based on the distance relationships between key points of the shoulder, elbow, and wrist in the dynamic skeleton, as well as the angle between the arm vector and the horizontal direction, the limb extension tendency is calculated.

5. The coal preparation plant security method based on intelligent video analytics according to claim 4, characterized in that, The method involves calculating the distance between the feet based on the positions of key points on both feet within the dynamic skeleton, and then combining this with the body's center of gravity movement speed to calculate the support stability index. Specifically, this includes: Obtain the coordinates of the left and right foot key points; Calculate the Euclidean distance between the key points of both feet as the foot spacing; Obtain the coordinate sequence of the human body's center of gravity in the preset adjacent frames of the current frame, calculate the average moving speed of the human body's center of gravity per unit time, and obtain the human body's center of gravity moving speed. The preset adjacent frames include the current frame and at least one frame before the current frame. The product of the time interval between video frames and the speed at which the human body's center of gravity moves is taken as the theoretical displacement of the human body's center of gravity within the time interval. The support stability index is obtained by dividing the distance between the feet by the theoretical displacement of the body's center of gravity during the time interval.

6. The coal preparation plant security method based on intelligent video analytics according to claim 4, characterized in that, The calculation of the center of gravity offset moment based on the horizontal offset distance between the key points of the neck and hips in the dynamic skeleton, the trunk length, and the trunk tilt angle specifically includes: Obtain the horizontal coordinates of the key points on the neck and the key points on the hip; Calculate the absolute value of the horizontal coordinate difference between the key points of the neck and the key points of the hip to obtain the horizontal offset distance; Obtain the Euclidean distance between the neck key point and the hip key point to get the torso length; Based on the positional relationship of key points of the torso in the dynamic skeleton, the angle between the center line of the torso and the vertical direction is calculated to obtain the torso tilt angle. Divide the horizontal offset distance by the torso length, and then multiply by the sine of the torso tilt angle to obtain the center of gravity offset moment.

7. The coal preparation plant security method based on intelligent video analytics according to claim 4, characterized in that, The calculation of limb extension tendency based on the distance relationships between key points of the shoulder, elbow, and wrist in the dynamic skeleton, as well as the angle between the arm vector and the horizontal direction, specifically includes: Obtain the coordinates of the shoulder key points, elbow key points, and wrist key points; Calculate the Euclidean distance between the shoulder key points and the wrist key points, and use it as the shoulder-wrist distance; Calculate the sum of the Euclidean distances between the shoulder key point and the elbow key point, and between the elbow key point and the wrist key point, as the total length of the arm zigzag line; Calculate the ratio of the shoulder-wrist distance to the total length of the arm's zigzag line to obtain the distance scaling factor; Obtain the cosine of the angle between the arm vector and the horizontal direction; Multiplying the distance scaling factor by the cosine value yields the limb extension tendency.

8. The coal preparation plant security method based on intelligent video analytics according to claim 1, characterized in that, The process of identifying the object's behavioral state and determining whether it is an illegal operation state based on changes in the support stability index, center of gravity offset moment, and limb extension tendency specifically includes: Obtain time-series data of the supporting stability index within a preset time window. When the data characteristics of the supporting stability index in the time-series data change from fluctuation to being consistently higher than its historical average, the object is determined to change from walking state to stationary state. Acquire the time series data of the center of gravity offset moment within a preset time window. When the data characteristics of the center of gravity offset moment in the time series data change from being maintained within the first historical reference range to being continuously higher than its first historical limit value, it is determined that the object's torso is continuously tilting forward toward the device side. Acquire time-series data of limb extension tendency within a preset time window. When the data characteristics of limb extension tendency in the time-series data change from within the second historical reference range to continuously exceeding its second historical limit value, and its upward trend overlaps with the upward trend of the center of gravity offset moment in time, it is determined that the person's arm has made a horizontal extension movement towards the equipment side. If all of the above judgment results are true within the preset time window, then the behavior state of the judgment object is an illegal operation state.

9. The coal preparation plant security method based on intelligent video analysis according to claim 1, characterized in that, The prediction of future position based on motion vectors of wrist key points in the dynamic skeleton, and the calculation of dynamic conflict time, specifically include: Obtain the normalized coordinates of the wrist key points in the current frame as the current position; Obtain the coordinate sequence of wrist key points over at least three consecutive frames; Calculate the instantaneous velocity vector of the wrist key points in the current frame based on the coordinate sequence; Calculate the instantaneous acceleration vector of the wrist key points in the current frame based on the coordinate sequence; Based on the current position, instantaneous velocity vector, and instantaneous acceleration vector, a uniformly accelerated motion model is constructed; The future positions of key points on the wrist are calculated using a uniformly accelerated motion model after a preset time period. Calculate the minimum distance between the future location and the boundary of the pre-defined danger zone; Divide the minimum distance by the component of the wrist's velocity along the direction pointing towards the danger zone to obtain the dynamic conflict time.

10. A security system for a coal preparation plant based on intelligent video analytics, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is executed by the processor, it implements the steps of the coal preparation plant security method for intelligent video analysis as described in any one of claims 1-9.

Images