Intelligent acquisition and analysis method and system based on big data model

Abstract

The invention discloses an intelligent acquisition and analysis method and system based on a big data model. The method comprises the following steps: 1, carrying out multi-modal identity verification and health state analysis; 2, construction safety real-time monitoring and behavior prediction are carried out; 3, special equipment dynamic security area modeling and risk early warning are carried out; and step 4, carrying out intelligent assessment on the leaving article loss risk. According to the invention, full-process closed-loop management and intelligent collaboration of construction site supervision can be realized, a full-process safety management control closed loop of entrance, on-duty and departure is constructed, a complete chain from risk early warning to disposal feedback is realized, a model dynamic optimization mechanism based on big data continuously improves the adaptability of the system to complex construction scenes, and the safety of construction sites is improved. And the manual intervention dependence is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data analysis, in particular to an intelligent collection and analysis method and system based on a big data model. BACKGROUND

[0002] The intelligent collection and analysis of the big data model refers to the efficient collection, fusion processing and deep mining of multi-source heterogeneous data, the construction of a self-adaptive learning model, the realization of an intelligent closed loop from data to decision, and the construction of a self-adaptive learning model.

[0003] The existing collection and analysis method and system have low intelligence, the content of detection and analysis is single, safety supervision is passive and lagging, and the use of the collection and analysis method and system is affected, therefore, the intelligent collection and analysis method and system based on the big data model are proposed. SUMMARY

[0004] In view of the defects in the prior art, the present application provides an intelligent collection and analysis method of a big data model, comprising the following steps:

[0005] Step one: multi-modal identity verification and health state analysis;

[0006] Step two: real-time monitoring of construction safety and behavior prediction;

[0007] Step three: dynamic security area modeling of special equipment and risk early warning;

[0008] Step four: intelligent evaluation of off-site article loss risk.

[0009] Further, the specific process of multi-modal identity verification and health state analysis is as follows:

[0010] An intelligent detection terminal is deployed at the entrance and exit of the construction site, and the following modules are integrated in the intelligent detection terminal:

[0011] A three-dimensional structured light face recognition module adopts a ToF (Time-of-Flight) sensor to collect face depth information, extracts a feature vector through a pre-trained ResNet-50 model, and performs cosine similarity matching with a permitted personnel access database (containing encrypted face features of employees and visitors), if the similarity is lower than a threshold value or no record is matched, an alarm is triggered;

[0012] A multi-spectral human body detection module collects body temperature data through an infrared thermal imager, and simultaneously performs alcohol concentration gas sensing detection to obtain the alcohol concentration of the human body;

[0013] The dynamic health threshold generation module generates a dynamic threshold interval based on a historical big data model (body temperature and alcohol concentration distribution of personnel of the same type in the past 30 days) through a Gaussian mixture model (GMM), and detects and analyzes the body temperature data and the alcohol concentration of the human body. If any one of the body temperature data and the alcohol concentration of the human body exceeds the dynamic threshold interval for two consecutive times, an alarm is triggered.

[0014] The human image acquisition module acquires human images, analyzes the gait characteristics (step frequency and limb swing amplitude) of the human images, and obtains human gait information.

[0015] The human gait information is compared and analyzed with a preset gait big data model. When the human gait information is abnormal, a secondary verification alarm is triggered, i.e., re-verification is required.

[0016] The process of obtaining human gait information is as follows:

[0017] Feature points of the human image are extracted, i.e., two ankle points, shoulder points, wrist points, hip points, and knee points are extracted.

[0018] The periodic displacement of the two ankle points in the continuation frame of the human image is analyzed to calculate the step frequency.

[0019] The step distance is obtained according to the difference between the maximum and minimum values of the distance between the two ankle points.

[0020] The horizontal displacement standard deviation of the shoulder points and wrist key points on both sides is calculated to obtain the arm swing amplitude.

[0021] The vertical motion trajectory variance of the hip points and knee points is calculated to obtain the leg swing amplitude.

[0022] The phase difference of the motion trajectories of the two ankle points is compared to obtain the gait symmetry parameter.

[0023] The step frequency, step distance, arm swing amplitude, leg swing amplitude, and gait symmetry parameter form the human gait information.

[0024] The step frequency, step distance, arm swing amplitude, and leg swing amplitude are introduced into the preset gait big data model for retrieval analysis. When at least two of the step frequency, step distance, arm swing amplitude, and leg swing amplitude are retrieved in the preset gait big data model, and the gait symmetry parameter is less than a preset value, it indicates that the gait information is normal.

[0025] When at least two of the step frequency, step distance, arm swing amplitude, and leg swing amplitude do not exist in the preset gait big data model, or the gait symmetry parameter is greater than or equal to the preset value, it indicates that the gait information is abnormal.

[0026] Further, the specific process of the construction safety real-time monitoring and behavior prediction is as follows:

[0027] Deploy a distributed intelligent camera network in the construction area, transmit 4K video streams in real time through the RTSP protocol, align multiple camera perspectives using the SIFT algorithm, and generate a panoramic construction scene three-dimensional point cloud map;

[0028] Then, safety equipment recognition is performed, an improved YOLOv7 model is used, an attention mechanism (CBAM module) is embedded at the input end of the YOLOv7 model, and safety helmets, reflective vests, and reflective vest color codes are recognized;

[0029] When a safety helmet is not recognized on any human body or a reflective vest is not recognized or the reflective vest color code does not match the type of work (such as an electrician not wearing a yellow reflective vest), a construction warning is triggered;

[0030] Then, regional compliance monitoring is performed, the construction area boundary is divided based on a U-Net segmentation model, the type of work and the mapping relationship between the type of work and the area are stored in a type of work label database, the worker's trajectory is tracked through a Kalman filter, and if the worker deviates from the specified area for more than a time T (T = 5 minutes), a regional violation report is generated;

[0031] Abnormal behavior detection: a spatio-temporal graph convolutional network (ST-GCN) is used to analyze consecutive video frames, run (speed > 2 m / s) and jump (frequency > 3 times / minute) behaviors are detected, and are compared with historical behavior database (stores typical motion patterns of types of work), when the number of detected runs or jumps deviates from the standard data in the historical behavior database by more than a preset value, it is determined to be abnormal, and a worker violation warning is triggered.

[0032] Further, the production building material stacking area is marked in the panoramic construction scene three-dimensional point cloud map, the building material stacking height is collected, the position of the building material with a stacking height exceeding the warning value is extracted, the position is marked as a warning building material position, the preset range of the position is marked as a safety warning area, the safety warning area is monitored by the intelligent cameras in the distributed intelligent camera network, when the intelligent cameras collect human body images outside the safety warning area, the human body images are processed, the face orientation and the body orientation of the human body images are collected;

[0033] The face orientation and the body orientation are collected f times, the number W1 of times that the face orientation points to the warning building material position and the number W2 of times that the body orientation points to the warning building material position are extracted;

[0034] When W1 and W2 are both greater than a preset value, and the human body moving speed is greater than a preset value, a worker prompt is triggered;

[0035] When the human position enters the safety warning area, the color coding of the reflective vest is analyzed to determine whether it is a person allowed to stay in the safety warning area. When it is determined that it is a person not allowed to stay in the safety warning area, a worker violation warning is triggered.

[0036] Further, the specific process of the special equipment dynamic security area modeling and risk early warning is as follows:

[0037] The special equipment includes a tower crane equipment. The device pose is reconstructed in real time through LiDAR point cloud data and visual SLAM algorithm, and a three-dimensional pose matrix of the tower crane equipment is generated, and a motion trajectory in the next 5 seconds is predicted.

[0038] Kinematic modeling is performed, and the motion equation of the tower crane equipment boom is as follows:

[0039]

[0040] θ(t) is the angle of the boom at time t, representing the predicted position at a certain time in the future;

[0041] θ0 is the initial angle of the boom, that is, the measured angle at the current time, which is obtained in real time through LiDAR point cloud matching or visual SLAM algorithm;

[0042] ω is the real-time angular velocity, α is the acceleration, and t is the time variable;

[0043] Trajectory prediction is performed, and an LSTM network is used to predict the motion trajectory in the next 5 seconds. The input is a historical pose sequence, and the output is a set of trajectory points {P future (t)}, t = 1 ~ 5 seconds;

[0044] The dynamic safety distance d is obtained through the formula

[0045] d0 is the static safety distance, ω is the real-time angular velocity of the device, which is obtained through the built-in IMU of the device, and ω max is the maximum allowed angular velocity of the device;

[0046] Taking the center of the tower crane equipment as the origin, a sector area S, that is, a dynamic security area, is generated according to the current angle θ of the boom, the swing angle range of the boom, and the dynamic safety distance d;

[0047] The sector area S is superimposed to the monitoring picture through the OpenGL rendering engine;

[0048] Then, personnel trajectory prediction is performed. The monitoring image is processed through the optical flow method. The Shi-Tomasi algorithm is used to detect the personnel contour corner points. Based on the Lucas-Kanade algorithm, the displacement vectors of the consecutive 5 frames are calculated to generate the personnel moving speed.​

[0049] Personnel trajectory prediction, fusion of optical flow data through Kalman filter, prediction of personnel position in future 2 seconds, through formula Get the personnel position in the next 2 seconds, P current The current position coordinates of personnel, υ avg Average moving speed, calculate the moving speed of personnel by optical flow method, α avg Average acceleration;

[0050] When , that is, it is determined that there is an entry risk, and a worker entry warning is triggered.

[0051] Further, the specific process of the intelligent assessment of the off-site article loss risk is as follows:

[0052] When multi-modal identity verification is performed, that is, the weight information G1 of the verified personnel is collected, and the corresponding type of work of the personnel is extracted, and the walking image of the personnel at the time of entry, that is, the personnel entry image, is collected;

[0053] When the personnel needs to leave the construction site, the weight information G2 of the personnel is collected again, and the walking image of the personnel walking towards the exit, that is, the personnel leaving image, is collected in advance, the corresponding type of work of the personnel is imported into the preset mapping set, and the weight increase standard coefficient b of the corresponding type of work at the time of leaving is retrieved;

[0054] Through the formula G2*b-G1=Gg, the evaluation difference Gg obtained, when the absolute value of the evaluation difference Gg is greater than the preset value threshold d1, the off-site article loss warning is triggered;

[0055] When the absolute value of Gg is greater than the threshold d2, the personnel entry image and the personnel leaving image are analyzed to determine whether there is a human body abnormality, and when there is a human body abnormality, the off-site article loss warning is triggered.

[0056] Further, the determination process of the human body abnormality is as follows:

[0057] First, bag / backpack recognition is performed on the personnel entry image and the personnel leaving image, when the personnel entry image and the personnel leaving image both recognize a bag, human body posture recognition is directly performed, when the human body posture is abnormal, it is determined that the human body is abnormal, and the off-site article loss warning is triggered;

[0058] When the personnel entry image does not recognize a bag / backpack, and the personnel leaving image recognizes a bag / backpack, a prompt is issued at this time, prompting the personnel to remove the bag / backpack before collecting the human body weight, which is marked as G3;

[0059] The difference between G3 and G1 is calculated to obtain a secondary evaluation difference, and when the secondary evaluation difference is greater than a preset value d2, it is determined that the human body is abnormal, that is, a departure loss warning is triggered.

[0060] Further, the specific process of human posture recognition is as follows: feature point extraction is performed on the personnel entering image, that is, two ankle points, two shoulder points, two elbow points and two wrist points are extracted;

[0061] The two ankle points are processed to obtain an entering step distance y1 and an entering step frequency y2.

[0062] The shoulder points, elbow points and wrist points on the same side are extracted, the shoulder points and wrist points are connected to obtain evaluation lines L1 and L2, the included angle U of the evaluation lines L1 and L2 is monitored, and then the difference between Umax and Umin is calculated to obtain an entering swing angle difference z1.

[0063] The same processing is performed on the personnel leaving image and the personnel entering image to obtain a leaving step distance y3, a leaving step frequency y4 and a leaving swing angle difference z2.

[0064] Then, the formula (|1-y3 / y1|)+(|1-y4 / y2|)+(|1-z2 / z1|)=yz is used to obtain the overall posture evaluation parameter yz, and when the overall posture evaluation parameter yz is greater than a preset value, it indicates that the human body is abnormal.

[0065] The intelligent collection and analysis system based on a big data model comprises:

[0066] The intelligent detection terminal is used for multi-modal identity verification and health state analysis.

[0067] The construction safety monitoring module is used for real-time monitoring and behavior prediction of construction safety.

[0068] The special equipment monitoring module is used for dynamic security area modeling and risk early warning of special equipment.

[0069] The departure loss evaluation module is used for intelligent evaluation of departure loss risk.

[0070] The beneficial effects of the present application are as follows:

[0071] The dual biological feature verification fusing three-dimensional face recognition and dynamic gait analysis significantly improves the reliability of identity impersonation detection and eliminates the risk of illegal personnel intrusion.

[0072] Through high-precision safety equipment recognition and work type matching mechanism, the detection ability of illegal behavior of not wearing safety equipment or inconsistent equipment is strengthened.

[0073] Combined with panoramic three-dimensional map and real-time trajectory tracking technology, the worker position compliance is dynamically monitored, and the misentry into high-risk areas is effectively prevented.

[0074] Based on the abnormal action prediction system of behavior pattern analysis, high-risk actions are early warned, and the risk of sudden accidents is reduced.

[0075] Through the bidirectional prediction mechanism of equipment motion trajectory and personnel position, the potential collision risk of equipment and personnel is early warned, the protection limit of traditional static safety area is broken through, the adaptive dynamic safety distance calculation model is calculated, the protection range is adjusted according to the real-time motion parameters of the equipment, and the safety blind area caused by high-speed motion of the equipment is eliminated.

[0076] Combined with weight change analysis and cross-period gait feature comparison, the hidden behavior of workers carrying illegal articles is effectively identified, the multi-modal data cross-validation mechanism is used to reduce the misjudgment caused by private articles interference, the verification accuracy is improved, the whole-process safety management control closed loop of entry, on-duty and exit is constructed, the complete chain from risk early warning to disposal feedback is realized, the model dynamic optimization mechanism based on big data is used to continuously improve the adaptability of the system to complex construction scenes, and the dependence on manual intervention is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0077] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the description of the specific embodiments or the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, each element or part is not necessarily drawn according to the actual proportion.

[0078] Figure 1 The flowchart of the present application. DETAILED DESCRIPTION

[0079] The embodiments of the technical solutions of the present application will be described in detail below with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present application, and therefore only serve as examples, but cannot limit the protection scope of the present application.

[0080] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in the present application should be understood as the usual meanings understood by the skilled in the art to which the present application belongs.

[0081] As Figure 1 shown, the intelligent collection and analysis method based on big data model includes the following steps:

[0082] Step one: multi-modal identity verification and health status analysis;

[0083] Step two: real-time monitoring of construction safety and behavior prediction;

[0084] Step three: dynamic security area modeling of special equipment and risk warning;

[0085] Step four: intelligent assessment of off-site article loss risk.

[0086] The specific process of multi-modal identity verification and health status analysis is as follows:

[0087] Intelligent detection terminals are deployed at the entrances and exits of the construction site, and the following modules are integrated in the intelligent detection terminals:

[0088] Three-dimensional structured light face recognition module: ToF (Time-of-Flight) sensor is used to collect face depth information, and a pre-trained ResNet-50 model is used to extract feature vectors. The cosine similarity is matched with the authorized personnel access database (including encrypted face features of employees and visitors). If the similarity is lower than the threshold or no record is matched, an alarm is triggered;

[0089] Multi-spectral human body detection module: infrared thermal imager is used to collect body temperature data, and alcohol concentration gas sensing detection is performed synchronously to obtain human alcohol concentration;

[0090] Dynamic health threshold generation module: based on historical big data model (body temperature and alcohol concentration distribution of personnel of the same type in the past 30 days), a dynamic threshold interval is generated by Gaussian Mixture Model (GMM). Body temperature data and human alcohol concentration are detected and analyzed. If any one of the two consecutive detection values exceeds the dynamic threshold interval, an alarm is triggered;

[0091] Human image acquisition module: human image acquisition is performed, and gait features (step frequency, limb swing amplitude) of human image are analyzed to obtain human gait information;

[0092] Compare and analyze human gait information with preset gait big data model. When human gait information is abnormal, a secondary verification alarm is triggered, i.e. re-verification is required;

[0093] The specific process of human gait information acquisition is as follows:

[0094] Feature points of human image are extracted, i.e. two ankle points, shoulder points, wrist points, hip points and knee points are extracted;

[0095] The periodic displacement of the two ankle points in the continuation frame of the human image is analyzed to calculate the step frequency;

[0096] According to the maximum and minimum difference of the distance between the two ankle points, the step distance is obtained;

[0097] The horizontal displacement standard deviation of the shoulder points and wrist key points on both sides is calculated to obtain the arm swing amplitude;

[0098] The vertical motion trajectory variance of the hip points and knee points is calculated to obtain the leg swing amplitude;

[0099] The phase difference of the motion trajectories of the two ankle points is compared to obtain the gait symmetry parameter;

[0100] Step frequency, step distance, arm swing amplitude, leg swing amplitude, and gait symmetry parameter form human gait information;

[0101] Step frequency, step distance, arm swing amplitude, and leg swing amplitude are introduced into the preset gait big data model for retrieval analysis. When at least two of step frequency, step distance, arm swing amplitude, and leg swing amplitude are retrieved in the preset gait big data model, and the gait symmetry parameter is less than the preset value, it means that the gait information is normal;

[0102] When at least two of step frequency, step distance, arm swing amplitude, and leg swing amplitude do not exist in the preset gait big data model, or the gait symmetry parameter is greater than or equal to the preset value, it means that the gait information is abnormal;

[0103] Through the above process, double biological feature cross verification is realized to enhance identity impersonation protection. Through the combination of three-dimensional face recognition and dynamic gait analysis, even if the attacker forges static face information (such as photo, mask), abnormal gait features can still trigger secondary verification, significantly improving the difficulty of identity impersonation.

[0104] Dynamic health threshold adapts to different types of work, reduces the false negative rate, and generates dynamic threshold based on historical data of different types of work, avoiding false judgment of health indicators due to differences in types of work (such as higher basic body temperature of high-temperature workers), and improving the rationality of monitoring.

[0105] Concealed health status monitoring reduces human interference. Through non-contact gait analysis, abnormal signs such as drunkenness and fatigue (such as unstable gait and abnormal limb swing amplitude) can be detected in a non-susceptible state, avoiding workers deliberately concealing problems;

[0106] If the attacker steals the face photo of a legitimate employee and tries to pass through three-dimensional face recognition. The system directly triggers an alarm because the photo cannot provide depth information (ToF sensor detection fails). Even if the attacker bypasses face detection, its gait features (such as limping and abnormal step frequency) do not match the preset model, and secondary verification intercepts again.

[0107] xxThe worker enters the construction site after drinking, and the alcohol concentration does not exceed the traditional fixed threshold (such as 60 mg / 100 mL), but the dynamic health threshold generates a lower warning value (such as 40 mg / 100 mL) based on historical data of electrician workers. At the same time, the gait analysis shows that the arm swing amplitude is reduced and the step length is irregular, and the system triggers an alarm after comprehensive judgment, preventing him from going to work.

[0108] xxThe worker's body temperature rises to 37.8℃ due to continuous work, but the dynamic threshold allows an upper limit of 38℃ based on data of the same type of workers. However, the gait analysis shows that the leg swing trajectory variance is abnormal (leg lifting weakness), and the system determines that it is in a state of fatigue, prompting the management personnel to arrange a rest.

[0109] The specific process of the construction safety real-time monitoring and behavior prediction is as follows:

[0110] Deploy a distributed intelligent camera network in the construction area, transmit 4K video stream in real time through RTSP protocol, align multiple camera perspectives using SIFT algorithm, and generate panoramic construction scene three-dimensional point cloud map;

[0111] Then perform safety equipment recognition, use improved YOLOv7 model, embed attention mechanism (CBAM module) at the input end of YOLOv7 model, recognize safety helmet, reflective vest and reflective vest color coding;

[0112] When safety helmet is not recognized on any human body or reflective vest is not recognized or reflective vest color coding does not match the type of work (such as electrician does not wear yellow reflective vest), trigger construction warning;

[0113] Then perform regional compliance monitoring, divide the construction area boundary based on U-Net segmentation model, combine with the work type label database (store the mapping relationship between work type and area), track the worker's trajectory through Kalman filter, and if the worker deviates from the specified area for more than time T (T = 5 minutes), generate a regional violation report;

[0114] Abnormal behavior detection: use spatio-temporal graph convolution network (ST-GCN) to analyze consecutive video frames, detect running (speed > 2 m / s) and jumping (frequency > 3 times / minute) behaviors, and compare with historical behavior database (store typical motion patterns of work type), when the number of detected running or jumping deviates from the standard data in the historical behavior database by more than a preset value, determine as abnormal, trigger worker violation warning;

[0115] Generate three-dimensional point cloud map through multi-camera cooperation, eliminate the blind area of traditional single-view monitoring, restore the overall situation of construction scene in real time, and ensure full coverage monitoring of dangerous areas such as high-altitude falling object area and equipment operation area.

[0116] Through the improved YOLOv7 model combined with the attention mechanism, the detection accuracy of the matching of safety helmets, reflective clothing colors, and types of work is improved, effectively intercepting illegal operations of not wearing equipment or equipment mismatch.

[0117] Based on the binding relationship between types of work and areas, the worker trajectory deviation is monitored in real time, such as electricians entering high-altitude work areas by mistake, timely warning of border crossing behavior, and preventing entering high-risk environments by mistake.

[0118] Through spatio-temporal behavior pattern analysis, high-risk actions that do not conform to the normal state of the type of work, such as frequent jumping and rapid running, are identified to provide early warning of potential accidents.

[0119] For example, a reinforcement worker enters a construction area without wearing a safety helmet. Distributed cameras locate his position through panoramic maps, and the improved YOLOv7 model accurately identifies that his head is not covered with a safety helmet, while detecting that the color of his reflective clothing (blue) does not match the standard color of the reinforcement worker (orange), immediately triggering a voice warning and notifying the supervisor.

[0120] A painter mistakenly enters the tower crane loading area to look for tools. The system tracks his trajectory through the U-Net segmented construction boundary and Kalman filter, finds that he has been staying in the unauthorized area for more than 5 minutes, automatically generates a violation report and sends it to the safety officer's handheld terminal, prompting timely removal.

[0121] In the scaffold operation area, a worker suddenly runs on a narrow platform (speed exceeds 2m / s) due to an emergency. The ST-GCN model analyzes the continuous action frames, and compares them with the historical database to find that the behavior deviates significantly from the standard action pattern of high-altitude platform operation, immediately triggering an alarm and requiring the worker to stop the action through the broadcast system.

[0122] Mark the production building material stacking area in the panoramic construction scene three-dimensional point cloud map, collect the building material stacking height, extract the position of the building material whose stacking height exceeds the warning value, mark it as a warning building material position, mark the preset range of the position as a safety warning area, and monitor the safety warning area through the intelligent camera in the distributed intelligent camera network. When the intelligent camera collects human body images outside the safety warning area, process the human body images, collect the face orientation and body orientation of the human body images;

[0123] Collect the face orientation f times and the body orientation f times, extract the number W1 of times that the face orientation points to the warning building material position and the number W2 of times that the body orientation points to the warning building material position;

[0124] When W1 and W2 are both greater than a preset value, and the human body moving speed is greater than a preset value, a worker prompt is triggered;

[0125] When the human position enters the safety warning area, the color coding of the reflective vest is analyzed to determine whether it is a person allowed to stay in the safety warning area. When it is determined that it is not a person allowed to stay in the safety warning area, a worker violation warning is triggered;

[0126] Based on the real-time monitoring of the building material pile height, the safety warning area range is automatically determined and adjusted to avoid the protection failure problem caused by the change of the pile due to the traditional fixed area mark, and the risk response flexibility is improved.

[0127] Through continuous analysis of the face and body orientation, the attention tendency of the personnel to the dangerous area (such as repeatedly observing the building material pile) is recognized, and the action intention is judged by the moving speed to realize the active protection of early warning without entering.

[0128] Different scenarios are distinguished between personnel inside and outside the safety warning area, and differential warning is triggered by combining identity permission and behavior characteristics to reduce false alarms.

[0129] If the hoisting operation height of the steel reinforcement pile area exceeds the threshold value, the system automatically marks the surrounding 3 meters as a safety warning area. When the worker carries tools outside the area, he repeatedly turns his head to watch the pile (the face orientation is collected 4 times out of 5 times pointing to the pile), and moves towards the pile at a faster speed. The system determines that it has the intention to approach, and immediately broadcasts a prompt through the nearby camera that the front is dangerous;

[0130] A visitor without a reflective vest mistakenly enters the safety warning area marked as a tower crane hoisting area. The system identifies that he has no color coding of the reflective vest through the camera, determines that he is an unauthorized person, immediately triggers a sound and light alarm and connects the gate lock to lock the exit, and sends positioning information to the security personnel.

[0131] The specific process of the dynamic security area modeling and risk warning of the special equipment is as follows:

[0132] The special equipment includes a tower crane device, which is reconstructed in real time by LiDAR point cloud data and visual SLAM algorithm, generates a three-dimensional pose matrix of the tower crane device, and predicts the motion trajectory in the next 5 seconds;

[0133] Kinematic modeling is performed, and the motion equation of the tower crane device boom is as follows:

[0134]

[0135] θ(t) is the angle of the boom at time t, representing the predicted position at a certain time in the future;

[0136] θ0 is the initial angle of the boom, i.e. the measured angle at the current time, which is obtained in real time by LiDAR point cloud matching or visual SLAM algorithm;

[0137] ω is the real-time angular velocity, a is the acceleration read from the device PLC, t is the time variable;

[0138] Trajectory prediction is performed, and an LSTM network is used to predict the motion trajectory for the next 5 seconds, the input is a historical pose sequence, and the output is a trajectory point set {P future (t)}, t = 1 ~ 5 seconds;

[0139] The dynamic safety distance d is obtained by the formula

[0140] d0 is the static safety distance, ω is the real-time angular velocity of the device, which is obtained by the built-in IMU of the device, ω max is the maximum allowed angular velocity of the device;

[0141] Taking the center of the tower crane device as the origin, according to the current angle θ of the boom, the swing angle range of the boom and the dynamic safety distance d, a sector area S is generated, which is the dynamic security area;

[0142] The sector area S is superimposed on the monitoring picture through the OpenGL rendering engine;

[0143] Then the personnel trajectory prediction is performed, the monitoring image is processed by the optical flow method, the personnel contour corner points are detected by using the Shi-Tomasi algorithm, the displacement vectors of the continuous 5 frames are calculated based on the Lucas-Kanade algorithm, and the personnel moving speed is generated;

[0144] The personnel trajectory prediction is performed, the optical flow data is fused by the Kalman filter, and the personnel position in the next 2 seconds is predicted, and the personnel position in the next 2 seconds is obtained by the formula P current is the current position coordinate of the personnel, υ avg is the average moving speed, the moving speed of the personnel is calculated by the optical flow method, a avg is the average acceleration;

[0145] When , it is determined that there is an entry risk, and the worker entry warning is triggered;

[0146] By reconstructing the pose of the special device in real time and predicting its motion trajectory, and by fusing the personnel moving trajectory optical flow analysis, the spatial conflict between the device and the personnel is predicted in advance, and the lag of the traditional static protection area is broken through.

[0147] According to the real-time motion parameters (such as angular velocity and acceleration) of the device, the safety distance is dynamically calculated, the safety area coverage is avoided due to high-speed motion of the device, and the blind area of protection is eliminated.

[0148] ​The dynamic safety area is superimposed on the monitoring picture in real time by a three-dimensional rendering engine, which intuitively shows the range of the dangerous area and improves the risk perception ability of the on-site personnel.

[0149] When the tower crane operator starts the lifting operation, the boom rotates at a high angular velocity. The system predicts the swing trajectory of the boom in the next 5 seconds through the LSTM network and generates a fan-shaped dynamic safety area. At the same time, the optical flow method detects that the XX worker is moving quickly in the direction of the boom swing, and the Kalman filter predicts that he will enter the dangerous area in 2 seconds. The system immediately triggers an audible and visual alarm, and the tower crane operator receives the prompt and pauses the operation to avoid a collision accident.

[0150] The tower crane needs to speed up the lifting due to an emergency task. The system automatically expands the dynamic safety area range in real time when it detects that the angular velocity exceeds the regular value, covering the additional risk range caused by the acceleration. When a patrol inspector who is unaware of the speed change approaches, the system triggers an alarm 10 seconds in advance through personnel trajectory prediction to prevent him from entering the dangerous area.

[0151] The specific process of the intelligent assessment of the risk of loss of items leaving the site is as follows:

[0152] During multi-modal identity verification, the weight information G1 of the verified personnel is collected, and the corresponding type of work of the personnel is extracted, and the walking image of the personnel when entering, i.e., the personnel entering image, is collected;

[0153] When the personnel need to leave the construction site, their weight information G2 is collected again, and the walking image of the personnel walking towards the exit, i.e., the personnel leaving image, is collected in advance, the corresponding type of work of the personnel is imported into the preset mapping set, and the weight increase standard coefficient b of the corresponding type of work when leaving is retrieved;

[0154] The evaluation difference Gg obtained through the formula G2*b-G1=Gg, when the absolute value of the evaluation difference Gg is greater than the preset threshold value d1, the leaving item loss warning is triggered. The absolute value is set to prevent workers from hiding heavy objects (such as stones, bricks, etc.) on their bodies before entering the site and replacing the hidden heavy objects with site materials such as cables when leaving;

[0155] When the absolute value of Gg is greater than the threshold value d2, the personnel entering image and the personnel leaving image are analyzed to determine whether there is a human body anomaly, and when there is a human body anomaly, the leaving item loss warning is triggered;

[0156] Through the difference between the entering and leaving weights, the evaluation threshold is dynamically corrected in combination with the type of work standard coefficient, effectively distinguishing between normal private property increase and decrease and the behavior of illegally carrying construction materials (such as reinforcing steel bars and cables);

[0157] Fusion body weight changes and cross-period gait feature comparison (entry / exit image analysis), avoid single data source misjudgment (such as worker carrying personal tools caused by weight fluctuations), improve detection reliability.

[0158] When an anomaly is detected, by requiring to remove the backpack after retesting the weight, both can quickly exclude interference factors, and avoid direct search inspection caused by controversy;

[0159] For example, steel reinforcement violation warning: the mason enters the scene with a body weight of G1=70kg (without carrying tools), and leaves the scene with a body weight of G2=74kg. The system calculates and evaluates the difference Gg=74×1.02-70=5.48kg according to the standard coefficient b=1.02 of its type (allowing to carry a small amount of personal tools).

[0160] Since the warning threshold d1=5kg, the system determines that the limit is triggered. Further analysis shows that the angle difference z2 of the swing gait increases by 40% compared to when it enters the scene, and it does not report the carrying of tools, and finally confirms that it carries 3 steel bars in the backpack (the actual weight gain is 4.2kg).

[0161] Tool false alarm exclusion:

[0162] The electrician enters the scene with a body weight of G1=68kg (carrying an electrician bag), and leaves the scene with a body weight of G2=73kg. The system calculates Gg=73×1.15-68=15.95kg according to the type coefficient b=1.15 (allowing to carry professional tools).

[0163] Although it exceeds the threshold d1=5kg, image analysis shows that it carries a new tool box when it leaves (the system marks it as allowed equipment), and the gait features are consistent with when it enters, so it is determined to be in compliance and released.

[0164] Hidden carrying secondary verification:

[0165] The handyman leaves the scene with a body weight of G2=72kg (G1=65kg when entering), and the evaluation difference Gg=72×1.0-65=7kg (exceeding the threshold d2=6kg). The system requires it to remove the backpack and retest the weight G3=66kg, and the secondary evaluation difference is =66-65=1kg (<d2).

[0166] But image analysis found that its waist stride was abnormally shortened by 20% when it left, and finally it was found that it was tied with copper wire (1.2kg) around its waist, triggering an alarm.

[0167] The determination process of the human body anomaly is as follows:

[0168] First, the bag / carrying bag of the personnel entering the image and the personnel leaving the image are identified, when the personnel entering the image and the personnel leaving the image are identified as carrying bags, the human body posture recognition is directly performed, when the human body posture is abnormal, it is determined that the human body is abnormal, that is, the loss of warning of the article leaving the scene is triggered;

[0169] When the bag is not identified in the entry image and is identified in the exit image, a prompt is given, prompting the person to remove the bag before collecting the body weight, which is marked as G3;

[0170] The difference between G3 and G1 is calculated to obtain the secondary evaluation difference, and when the secondary evaluation difference is greater than the preset value d2, it is determined that the human body is abnormal, that is, the off-site article loss warning is triggered.

[0171] The specific process of human body posture recognition is as follows: feature point extraction is performed on the entry image of the person, that is, two ankle points, two shoulder points, two elbow points and two wrist points are extracted;

[0172] The two ankle points are processed to obtain the entry step distance y1 and the entry step frequency y2;

[0173] The shoulder point, elbow point and wrist point of the same side are extracted, the shoulder point and wrist point are connected to obtain evaluation lines L1 and L2, the included angle U of the evaluation lines L1 and L2 is monitored, and then the difference between Umax and Umin is calculated to obtain the entry swing included angle difference z1;

[0174] The same processing is performed on the exit image of the person and the entry image of the person to obtain the exit step distance y3, the exit step frequency y4 and the exit swing included angle difference z2;

[0175] Then, through the formula (|1-y3 / y1|)+(|1-y4 / y2|)+(|1-z2 / z1|)=yz, the overall posture evaluation parameter yz is obtained, and when the overall posture evaluation parameter yz is greater than the preset value, it indicates that the human body is abnormal;

[0176] Through dynamic comparison of gait features (step distance, step frequency, limb swing amplitude, etc.) of entry and exit images, even if the carried article does not significantly change the body weight (such as small volume metal parts), the irregular behavior can also be accurately identified through gait coordination abnormality (such as limb stiffness, reduced swing amplitude), and the detection sensitivity is improved;

[0177] The gait data of the same person entering is taken as the individualized reference, rather than the general model, which avoids false positives caused by congenital gait differences, and reduces the misjudgment rate;

[0178] Based on visual feature analysis, the irregular detection is realized without searching or opening the bag for inspection, which not only guarantees the dignity of workers, but also improves the passing efficiency.

[0179] For example, the welder XX worker enters with a step distance of 70 cm and an arm swing amplitude of 25°, and when leaving, the step distance is shortened to 60 cm (step distance change rate 14.3%) due to hiding metal parts in the trouser pocket, and the arm swing amplitude is reduced to 18° (swing amplitude difference 28%). The system calculates the overall posture evaluation parameter yz = 0.14 + 0.28 + 0.15 = 0.57, which exceeds the threshold value 0.5, triggering an alarm. After inspection, it is found that he violated the regulation by carrying welding rods.

[0180] The electrician XX worker enters with a step distance of 65 cm due to carrying a tool box, and the step distance is shortened to 60 cm when leaving due to the addition of tools. The system finds that the difference in arm swing amplitude between entering and leaving is only 5%, and the overall parameter yz = 0.08 (<0.5), which is determined to be in compliance with the tool carrying, avoiding false positives.

[0181] The handyman XX worker hides copper screws (total weight 0.8 kg) in his shoes, and when leaving, the gait symmetry parameter is abnormal. Although the body weight change does not exceed the threshold value, yz = 0.62 triggers a secondary check, and the illegal items are finally found.

[0182] The intelligent collection and analysis system based on big data model includes:

[0183] The intelligent detection terminal is used for multi-modal identity verification and health status analysis.

[0184] The construction safety monitoring module is used for real-time monitoring and behavior prediction of construction safety.

[0185] The special equipment monitoring module is used for dynamic security area modeling and risk warning of special equipment.

[0186] The off-site loss evaluation module is used for intelligent evaluation of off-site loss risk.

[0187] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and are not limited thereto; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the specification of the present application.

Claims

1. An intelligent collection and analysis method based on a big data model, characterized in that: The following steps are involved: Step 1: Perform multimodal identity authentication and health status analysis; Step 2: Real-time monitoring of construction safety and behavior prediction; Step 3: Dynamic security area modeling and risk warning for special equipment; Step 4: Conduct intelligent assessment of the risk of loss of items leaving the site.

2. The intelligent data collection and analysis method based on a big data model according to claim 1 is characterized in that: The specific process of multimodal identity authentication and health status analysis is as follows: Smart detection terminals are deployed at the entrances and exits of the construction site, integrating the following modules: The 3D structured light face recognition module uses a ToF sensor to collect facial depth information, extracts feature vectors using a pre-trained ResNet-50 model, and performs cosine similarity matching with a database of authorized personnel. If the similarity falls below a threshold or no record is matched, an alert is triggered. The multi-spectral human body detection module collects body temperature data through an infrared thermal imager and simultaneously performs alcohol concentration gas sensor detection to obtain the human alcohol concentration; Dynamic health threshold generation module: Based on the historical big data model, the module generates dynamic threshold intervals through the Gaussian mixture model, and detects and analyzes body temperature data and human alcohol concentration. If either the body temperature data or the human alcohol concentration exceeds the dynamic threshold interval for two consecutive times, an alert is triggered. The human body image acquisition module collects human body images, performs gait feature analysis on the human body images, and obtains human gait information; The human gait information is compared and analyzed with the preset gait big data model. When there is an abnormality in the human gait information, a secondary verification alert is triggered, and re-verification is required.

3. The intelligent data collection and analysis method based on a big data model according to claim 1, characterized in that: The specific process of real-time construction safety monitoring and behavior prediction is as follows: Deploy a distributed intelligent camera network in the construction area, transmit 4K video streams in real time via the RTSP protocol, and use the SIFT algorithm to align multiple camera perspectives to generate a panoramic 3D point cloud map of the construction scene. Next, we use an improved YOLOv7 model to identify safety equipment. We embed an attention mechanism at the input of the YOLOv7 model to identify helmets, reflective clothing, and their color codes. When a helmet or reflective clothing is not recognized on any person, or the color coding of the reflective clothing does not match the type of work, a construction alert is triggered; Regional compliance monitoring is then performed. The construction area boundaries are demarcated based on the U-Net segmentation model. Combined with the work type label database, the worker trajectories are tracked using a Kalman filter. If a worker deviates from the designated area for more than T, a regional violation report is generated. Abnormal behavior detection: A spatiotemporal graph convolutional network is used to analyze continuous video frames to detect running and jumping behaviors, and then compare them with the historical behavior database. When the detected number of running or jumping times deviates from the standard data in the historical behavior database by more than a preset value, it is judged as abnormal and triggers a worker violation warning.

4. The intelligent data collection and analysis method based on a big data model according to claim 3 is characterized by: Mark the production building material stacking area in the panoramic construction scene 3D point cloud map, collect the building material stacking height, extract the building material location where the stacking height exceeds the warning value, mark it as the warning building material location, and mark the preset range of this location as the safety warning area. Monitor the safety warning area through the smart cameras in the distributed smart camera network. When the smart camera captures a human image outside the safety warning area, the human image is processed to collect the facial and torso orientation of the human image; Continuously collect f facial orientations and f body orientations, extract the number of times W1 when the facial orientation points to the warning building material location and the number of times W2 when the body orientation points to the warning building material location; When both W1 and W2 are greater than the preset values, and the human body movement speed is greater than the preset value, the worker prompt is triggered; When a person enters the safety warning area, the color code of his / her reflective clothing is analyzed to determine whether he / she is allowed to stay in the safety warning area. If he / she is determined to be not allowed to stay in the safety warning area, a worker violation warning is triggered.

5. The intelligent data collection and analysis method based on a big data model according to claim 1 is characterized in that: The specific process of dynamic security area modeling and risk warning for special equipment is as follows: Special equipment, including tower cranes, uses LiDAR point cloud data and visual SLAM algorithms to reconstruct the equipment's posture in real time, generate a three-dimensional posture matrix for the tower crane, and predict its motion trajectory within the next 5 seconds. For kinematic modeling, the motion equation of the tower crane arm is as follows: θ(t) is the angle of the boom at time t, indicating the predicted position at a certain moment in the future; θ0 is the initial angle of the boom, that is, the measured angle at the current moment, which is obtained in real time through LiDAR point cloud matching or visual SLAM algorithm; ω is the real-time angular velocity, α is the acceleration read from the device PLC, and t is the time variable; Perform trajectory prediction and use LSTM network to predict the motion trajectory for the next 5 seconds. The input is the historical pose sequence and the output is the trajectory point set {P future (t)}, t = 1 to 5 seconds; By formula Get the dynamic safety distance d; d0 is the static safety distance, ω is the real-time angular velocity of the device, which is obtained through the built-in IMU of the device. max is the maximum allowable angular velocity of the device; With the center of the tower crane as the origin, a fan-shaped area S, i.e., a dynamic security area, is generated based on the current boom angle θ, the boom swing angle range, and the dynamic safety distance d. Through the OpenGL rendering engine, the sector area S is superimposed on the monitoring screen; After that, we predict the trajectory of people. We process the surveillance images using the optical flow method. We input the surveillance images, use the Shi-Tomasi algorithm to detect the corner points of the people's contours, and calculate the displacement vectors of 5 consecutive frames based on the Lucas-Kanade algorithm to generate the movement speed of the people. To predict the trajectory of personnel, the Kalman filter is used to fuse the optical flow data to predict the position of personnel in the next 2 seconds. Get the personnel position in the next 2 seconds, P current is the current position coordinate of the personnel, υ avg Average moving speed, the moving speed of people in continuous frames is calculated by optical flow method, α avg is the average acceleration; when When the worker enters the building, it is determined that there is an entry risk and a worker entry warning is triggered.

6. The intelligent data collection and analysis method based on a big data model according to claim 1 is characterized by: The specific process of intelligent assessment of the risk of loss of items leaving the site is as follows: During multimodal identity verification, the weight information G1 of the person being verified is collected, and the type of work corresponding to the person is extracted. At the same time, the walking image of the person entering the user is collected, i.e., the person entering the user image. When a worker needs to leave the construction site, their weight information G2 is collected again. A walking image of the worker walking towards the exit is also captured in advance, i.e., the worker's leaving image. The worker's corresponding job type is imported into a preset mapping set, and the standard weight gain coefficient b of the corresponding job type when leaving is retrieved. The evaluation difference Gg is obtained by the formula G2*b-G1=Gg. When the absolute value of the evaluation difference Gg is greater than the preset threshold value d1, an off-site item loss warning is triggered; When the absolute value of Gg is greater than the threshold d2, the images of people entering and leaving are analyzed to determine whether there is any human body abnormality. If there is any human body abnormality, an alarm for lost items will be triggered.

7. The intelligent data collection and analysis method based on a big data model according to claim 6, characterized in that: The process of determining human body abnormality is as follows: First, shoulder bag / backpack recognition is performed on the images of people entering and leaving. If a shoulder bag is recognized in both the images, human posture recognition is directly performed. If there is any abnormality in the human posture, it is determined that the person is abnormal, which triggers an alarm for lost items. When a person enters the image but the bag / backpack is not recognized, and leaves the image and the bag / backpack is recognized, a prompt is issued to remind the person to remove the bag / backpack before collecting the body weight, and the person is marked as G3; The difference between G3 and G1 is calculated to obtain the secondary assessment difference. When the secondary assessment difference is greater than the preset value d2, it is determined that the human body is abnormal, which triggers an alarm for the loss of items at the exit.

8. The intelligent data collection and analysis method based on a big data model according to claim 7 is characterized in that: The specific process of human posture recognition is as follows: feature points are extracted from the image of the person entering, that is, two ankle points, two shoulder points, two elbow points and two wrist points are extracted; Process the two ankle points to obtain the entry stride length y1 and entry stride frequency y2; Extract the shoulder, elbow, and wrist points on the same side, connect the shoulder and wrist points to obtain the evaluation lines L1 and L2, respectively, monitor the angle U between the evaluation lines L1 and L2, and then calculate the difference between Umax and Umin to obtain the swing angle difference z1; The image of the person leaving is processed in the same way as the image of the person entering, and the leaving stride length y3, leaving stride frequency y4 and leaving swing angle difference z2 are obtained; Then, the overall posture evaluation parameter yz is obtained through the formula (|1-y3 / y1|)+(|1-y4 / y2|)+(|1-z2 / z1|)=yz. When the overall posture evaluation parameter yz obtained is greater than the preset value, it means that the human body is abnormal.

9. An intelligent data collection and analysis system based on a big data model, the analysis system being based on the analysis method according to any one of claims 1 to 8, characterized in that: include: Intelligent detection terminals for multimodal identity authentication and health status analysis; Construction safety monitoring module, used for real-time monitoring of construction safety and behavior prediction; Special equipment monitoring module, used for dynamic security area modeling and risk warning of special equipment; The off-site loss assessment module is used to intelligently assess the risk of off-site item loss.

Images