Intelligent remote monitoring and recognition system for highway construction safety behaviors

Abstract

The invention relates to the technical field of remote monitoring, in particular to a highway construction safety behavior intelligent remote monitoring recognition system which comprises a dispatching management module, a center detection module, a path screening module, an area monitoring module and a low-illumination monitoring module. According to the method, the priority transmission of key image data is realized by extracting the time stamp of the image data frame, the definition level and the area priority and combining the bandwidth dynamic scheduling, and the identification of the angle deviation of the camera is realized by identifying the static traffic facility, marking the image center point and recording the center track change. According to the method, the electronic fence and personnel track recognition are combined, a border crossing behavior path is extracted, the border crossing recognition continuity is enhanced, non-protection crossing detection of a dangerous area is achieved by extracting personnel distance changes and protection states, construction risk early warning coverage is optimized, and the night monitoring abnormity capture rate is increased through night light reflection track coherence and space comparison.

Description

Technical Field

[0001] The present invention relates to the technical field of remote monitoring, and particularly to an intelligent remote monitoring and identification system for highway construction safety behaviors. Background Art

[0002] The technical field of remote monitoring includes systems and methods for real-time or non-real-time information collection, transmission, processing, and display of a target area through a communication network. The core content of this technical field includes data acquisition of image acquisition devices, remote communication of transmission modules, information parsing and instruction control of a central processing system, real-time monitoring and playback support of a display terminal, covering video monitoring systems, remote control systems, data acquisition terminals, network transmission modules, and background server processing systems. The aim is to transmit on-site images, audio, or sensing data to a central control platform in real time through wired or wireless communication methods to achieve monitoring and management of the environment, personnel behaviors, equipment operating states, etc. of a remote target area, and it is applied to various scenarios such as urban traffic, public safety, industrial production, and infrastructure construction.

[0003] Among them, an intelligent remote monitoring and identification system for highway construction safety behaviors refers to a system applied to highway construction sites, which obtains activity images of construction workers in the construction area through an image acquisition unit, uses a behavior recognition method based on feature extraction to automatically identify specific actions, area access, and dangerous behaviors of construction workers, and combines a wireless network to transmit the recognition results and image data to a remote monitoring center in real time. The monitoring center classifies and processes the data according to set rules and records them. The technical matters covered include the layout methods of fixed or mobile camera devices, behavior pattern modeling methods based on human form feature analysis, behavior recognition processes based on time series analysis, data upload mechanisms achieved through network control, and instruction feedback mechanisms of the remote monitoring center, which are completed by image acquisition, local feature analysis, template matching recognition, wireless data transmission, and classification recording in the monitoring center.

[0004] Traditional highway construction safety behavior monitoring technologies cannot dynamically adjust transmission resources according to the importance and urgency of image content, resulting in possible delays or losses of key behavior images under high-load conditions, reducing response efficiency. There is a lack of continuous tracking and detection of the stability of camera position information. When the camera is offset due to environmental vibration or human interference, it cannot be recognized and adjusted in time, causing monitoring blind spots or image offsets. The monitoring of personnel behaviors relies on static image recognition and it is difficult to conduct continuous behavior screening in combination with the evolution process of personnel dynamic trajectories. In particular, behaviors approaching dangerous areas without protection are easily missed. There is a lack of a special recognition mechanism for reflective sign trajectories at night or in low-light environments, resulting in abnormal trajectory activities being difficult to capture in low-light scenarios, having a night monitoring blind spot, leading to monitoring failure and safety risks. Summary of the Invention

[0005] The object of the present invention is to solve the drawbacks existing in the prior art, and to propose an intelligent remote monitoring and identification system for highway construction safety behaviors.

[0006] To achieve the above object, the present invention adopts the following technical solutions: An intelligent remote monitoring and identification system for highway construction safety behaviors includes: The scheduling management module acquires the monitored image data frames, extracts the time stamp, picture clarity level, and area priority label of each frame, evaluates the urgency and importance of the image frames, combines the available bandwidth level of the node, adjusts the bandwidth resource allocation parameters of each image frame, and generates a data transmission scheduling configuration; The central detection module calls the data transmission scheduling configuration, analyzes the image frames and identifies static traffic facilities and fixed signs, obtains the center point of the image, identifies the change of the center point trajectory of consecutive frames, identifies the deviation risk of the camera angle, and generates a camera offset value; The path screening module calls the camera offset value, identifies the personnel movement trajectory by analyzing the monitored image, combines the boundary data of the electronic fence, records the number of fence crossings, crossing direction, and crossing time points, extracts the crossing trajectory pointing to the non-operation area, and generates an out-of-bounds behavior identification record; The area monitoring module obtains the continuous position information of the personnel according to the out-of-bounds behavior identification record, extracts the distance change sequence between the current position and the boundary of the dangerous area, combines the identification result of the wearing status of the protective equipment, detects the dangerous area crossing behavior under the condition of no protection, and generates a safety behavior identification record.

[0007] As a further solution of the present invention, the data transmission scheduling configuration includes image frame transmission priority sorting information, bandwidth resource allocation ratio data, and node signal load status records. The camera offset value is specifically the center point trajectory offset amplitude, the center point offset direction change amount, and the camera attitude offset angle. The out-of-bounds behavior identification record includes the fence crossing times statistics, the non-operation area pointing trajectory sequence, and the crossing event time stamp list. The safety behavior identification record is specifically the fast crossing action determination result, the no-protection status identification label, and the dangerous area approaching speed change curve.

[0008] As a further solution of the present invention, the scheduling management module includes: The image attribute extraction sub-module acquires the monitored image data frames, extracts the time stamp parameter, picture clarity level, and area priority label of each frame image, and generates image attribute information; The image frame priority evaluation sub-module calls the image attribute information, and evaluates the urgency value and importance level value of each image frame according to the area priority label and the picture clarity level, and generates an image frame importance evaluation index; The bandwidth resource dynamic configuration sub-module calls the image frame importance evaluation index, combines the available bandwidth level parameter of the node, extracts the node data transmission demand parameter, the node remaining bandwidth capacity parameter and the node load allocation status parameter, calculates the image frame bandwidth allocation adjustment amount, and obtains the data transmission scheduling configuration.

[0009] As a further solution of the present invention, the center detection module includes: The image center extraction sub-module obtains the data transmission scheduling configuration, collects the monitoring image frame data, analyzes the image frame and identifies the traffic facility image data and the fixed sign image data, obtains the center point in each frame of the image, and generates an image center point coordinate set; The trajectory change extraction sub-module calls the image center point coordinate set, extracts the horizontal displacement amount and the vertical displacement amount of the center point of the continuously collected image frames, calculates the adjacent frame center point displacement increment data, records the continuous displacement increment set, and generates the center point trajectory change amount; The camera offset evaluation sub-module calls the center point trajectory change amount, extracts the time series distribution of the continuous displacement increment set, calculates the camera offset level value, and compares it with the preset camera offset level interval to generate the camera offset amount value.

[0010] As a further solution of the present invention, the path screening module includes: The trajectory point extraction sub-module calls the camera offset amount value, collects the monitoring image frame data, identifies the contour of the construction workers in the image, extracts the position coordinate points of the construction workers in each frame, arranges the position coordinate point set in the order of time stamps, and generates a personnel trajectory data set; The crossing event extraction sub-module calls the personnel trajectory data set, obtains the electronic fence boundary data of the operation area, detects the spatial relationship state between each position coordinate point and the electronic fence boundary, detects the boundary crossing event, and records the trajectory points, crossing directions and crossing time points where the boundary crossing occurs, and generates a fence crossing event data set; The out-of-bounds trajectory screening sub-module calls the fence crossing event data set, compares according to the crossing direction and the operation area outer boundary direction data, extracts the trajectory segments whose crossing directions point to non-operation areas, and generates an out-of-bounds behavior recognition record.

[0011] As a further solution of the present invention, the area monitoring module includes: The position trajectory extraction sub-module obtains the out-of-bounds behavior recognition record, identifies the dangerous area signs in the highway construction area, sets the spatial boundary of the dangerous area, and obtains the dangerous area boundary data set; The dangerous approach screening sub-module calls the dangerous area boundary data set, extracts the distance value sequence and the corresponding time sequence between the current position coordinate point of the person and the dangerous area boundary point, identifies the continuous time interval and the position change amount, calculates the approaching speed risk score, and obtains the dangerous approach behavior detection record by comparing with the preset dangerous area rapid approach risk threshold; The unprotected crossing identification sub-module calls the dangerous approach behavior detection record, real-time detects the wearing status of the personal protective equipment of the construction personnel, identifies and extracts the dangerous area crossing behavior in the unprotected state, and generates a safety behavior identification record.

[0012] As a further solution of the present invention, the system further includes: The low-light monitoring module calls the safety behavior identification record, obtains the image data in the low-illumination environment at night and identifies the reflective signs in the image, extracts the trajectory path of the reflective bright spots in the continuous frames, analyzes the coherence of the trajectory direction and the interruption frequency, identifies the abnormal trajectory segments and compares the spatial relationship with the dangerous area boundary, and detects the abnormal movement trajectory at night to generate a night behavior monitoring record; The night behavior monitoring record specifically refers to the night trajectory coherence score, the abnormal trajectory offset distance, and the distribution range of the abnormal trajectory in the dangerous area.

[0013] As a further solution of the present invention, the low-light monitoring module includes: The reflective trajectory extraction sub-module calls the safety behavior identification record, obtains the image data in the low-illumination environment at night and identifies the reflective signs in the image, extracts the position coordinates of the reflective bright spots in the continuous frame images, records the trajectory path of the reflective bright spots under the continuous time sequence, and generates a reflective trajectory path data set; The trajectory anomaly identification sub-module calls the reflective trajectory path data set, analyzes the coherence of the change direction between the trajectory points and the trajectory interruption frequency, detects the abnormal trajectory segments, and generates a set of abnormal trajectory segments; The abnormal trajectory space detection sub-module calls the set of abnormal trajectory segments, combines the dangerous area boundary data, detects the spatial interaction relationship between the abnormal trajectory segments and the dangerous area boundary, detects the abnormal movement trajectory at night, and generates a night behavior monitoring record.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by extracting the image data frame timestamp, clarity level, and regional priority, and combining with dynamic bandwidth scheduling, the priority transmission of key image data is achieved. By identifying the center point of the static traffic facility marker image and recording the change of the center trajectory, the recognition of the camera angle offset is realized. By combining the electronic fence and the personnel trajectory recognition, the path of the crossing behavior is extracted to enhance the continuity of the crossing recognition. By extracting the change of the personnel distance and the protection status, the detection of unprotected crossing in the danger zone is realized, and the coverage of the construction risk warning is optimized. By the coherence and spatial contrast of the night-time reflective trajectory, the abnormal capture of night-time monitoring is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is the system flow chart of the present invention; Figure 2 is the flow chart of the scheduling management module of the present invention; Figure 3 is the flow chart of the center detection module of the present invention; Figure 4 is the flow chart of the path screening module of the present invention; Figure 5 is the flow chart of the area monitoring module of the present invention; Figure 6 is the flow chart of the low-light monitoring module of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0017] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more unless otherwise specifically defined.

[0018] Please refer to Figure 1 , an intelligent remote monitoring and recognition system for highway construction safety behaviors includes: The scheduling management module obtains the monitored image data frames, extracts the time stamps, picture clarity levels, and area priority labels of each frame, evaluates the urgency and importance of the image frames, and adjusts the bandwidth resource allocation parameters for each image frame in combination with the available bandwidth level of the node to generate a data transmission scheduling configuration; The central detection module calls the data transmission scheduling configuration, analyzes the image frames to identify static traffic facilities and fixed signs, obtains the center point of the image, identifies the change in the center point trajectory of consecutive frames, identifies the offset risk of the camera angle, and generates a camera offset value; The path screening module calls the camera offset value, identifies the movement trajectory of personnel by analyzing the monitored images, combines the boundary data of the electronic fence, records the number of fence crossings, crossing directions, and crossing time points, extracts the crossing trajectories pointing to non-operation areas, and generates an out-of-bounds behavior recognition record; The area monitoring module obtains the continuous position information of personnel according to the out-of-bounds behavior recognition record, extracts the distance change sequence between the current position and the boundary of the dangerous area, and combines the recognition result of the wearing status of protective equipment to detect the behavior of crossing the dangerous area without protection conditions and generate a safety behavior recognition record; The low-light monitoring module calls the safety behavior recognition record, obtains the image data in the low-illuminance environment at night and identifies the reflective signs in the image, extracts the trajectory path of the reflective bright spots in consecutive frames, analyzes the coherence of the trajectory direction and the interruption frequency, identifies abnormal trajectory segments and compares the spatial relationship with the boundary of the dangerous area to detect abnormal movement trajectories at night and generate a night behavior monitoring record.

[0019] The night behavior monitoring record specifically refers to the night trajectory coherence score, abnormal trajectory offset distance, and the distribution range of abnormal trajectories in the dangerous area.

[0020] Please refer to Figure 2 , the scheduling management module includes: The image attribute extraction sub-module obtains the monitored image data frames, extracts the time stamp parameters, picture clarity levels, and area priority labels of each frame of image, and generates image attribute information; First, based on the acquisition output of the monitoring device, the capture time information corresponding to each frame of the image is extracted as the timestamp parameter. The sampling frequency is set to 10 frames per second, so the time interval recorded for each frame is 0.1 seconds. For example, in the first frame captured, the timestamp is 12:00:00.0, and for the second frame, it is 12:00:00.1, and the time information is arranged in sequence accordingly. Subsequently, the picture clarity level parameter is extracted from the encoded metadata of the image frame. The clarity level is set to three levels, namely high definition (HD), standard definition (SD), and low definition (LD), which are encoded as 0, 1, and 2 respectively. In the example, the clarity encoding of the first frame is 0, and that of the second frame is 1, and the clarity level is recorded in this way; further, the preset area priority label in the construction area is extracted. The area is divided into high-priority area, medium-priority area, and low-priority area according to the construction importance, and the values are assigned as 3, 2, and 1 respectively. For example, the first frame located in the high-priority area is marked as 3, and the second frame located in the medium-priority area is marked as 2. All the extracted parameters are used to construct the original data set in the form of three columns: timestamp, clarity encoding, and priority encoding. See the following table for details: Table 1 Initial extraction table of image attributes: ; As shown in Table 1, the basic attribute information of the image frame is completed in terms of acquisition and preliminary collection, and all subsequent processing procedures are carried out based on the data in this table, and finally the image attribute information is generated.

[0021] The image frame priority evaluation sub-module calls the image attribute information, and based on the area priority label and the picture clarity level, evaluates the urgency value and importance level value of each image frame, and generates the image frame importance evaluation index; First, based on the area priority label data extracted from the image attribute information, the initial value of the urgency level is set according to the priority level. The high-priority area is assigned an urgency level of 3, the medium-priority area is 2, and the low-priority area is 1. The urgency value is directly mapped from the area priority label. For example, if the area priority of the first frame is 3, the urgency level is assigned as 3; subsequently, according to the picture clarity level, the clarity importance benchmark is set. The high definition is set to 1.0, the standard definition is set to 0.7, and the low definition is set to 0.5. For example, if the clarity of the first frame is 0 (HD), the importance is 1.0, and the clarity of the second frame is 1 (SD), the importance is 0.7. The urgency value and the picture clarity importance value are weighted and averaged to form the image frame importance evaluation index. For example, if the weight is set such that the area priority accounts for 70% and the clarity accounts for 30%, the importance evaluation index of the first frame is 2.4. In this way, the importance evaluation index of each frame of the image is calculated and archived, and finally the image frame importance evaluation index is generated.

[0022] The dynamic bandwidth resource configuration sub-module calls the image frame importance evaluation index, combines the available bandwidth level parameter of the node, extracts the node data transmission demand parameter, the node remaining bandwidth capacity parameter and the node load allocation status parameter, and uses the formula: ; Calculate the bandwidth allocation adjustment amount of the image frame and obtain the data transmission scheduling configuration; Among them, represents the bandwidth allocation adjustment amount of the image frame, represents the image frame importance evaluation index, represents the area priority label parameter, represents the current load level parameter of the node, represents the node remaining bandwidth capacity parameter, represents the current load bandwidth amount parameter of the node; First, set the total available bandwidth of the node to 100 Mbps. The node data transmission demand is calculated by dividing the total sum of the image frame sizes by the time period length. For example, if there are 10 frames with a size of 1 MB each, the demand is 80 Mbps. The node remaining bandwidth capacity is the total bandwidth of the current node minus the occupied bandwidth. For example, the remaining capacity is 20 Mbps. The node load allocation status is recorded as the ratio of the allocated bandwidth to the total bandwidth. For example, if 80% is occupied, the load status is 0.8. Use the following formula: ; Set the first frame , , , , , and substitute into the operation: ; That is, the first frame is allocated 22.998 Mbps of bandwidth. Calculate the bandwidth adjustment amount for all frames in turn according to this formula, and finally obtain the data transmission scheduling configuration. Among them, the bandwidth allocation adjustment amount of the image frame refers to the recommended bandwidth resource allocation amount dynamically calculated for each captured image frame after comprehensively considering the content value of the image frame, the area task priority, the available bandwidth resources of the node, and the current load status of the node. This parameter directly guides how much actual bandwidth resources are allocated to the image frame data packet in the remote transmission link. Its unit is Mbps. The larger the value, the more real-time link resources the image frame will occupy during transmission and the higher the priority for transmission guarantee. The smaller the value, the more delayed the image frame will be processed in the network resource scheduling. The specific effect is to dynamically distinguish different transmission priorities and rates according to the video data of different regions and different important tasks at the construction site, improve the remote monitoring clarity, real-time performance and continuous tracking ability of key construction behaviors and dangerous area operation behaviors, and at the same time optimize the overall bandwidth utilization rate of the node, reduce the risk of network transmission congestion and data frame loss.

[0023] Please refer to Figure 3 , the center detection module includes: The image center extraction sub-module obtains the data transmission scheduling configuration, collects the monitoring image frame data, analyzes the image frames and identifies the traffic facility image data and the fixed sign image data, obtains the center points in each frame of the image, and generates an image center point coordinate set; First, call the bandwidth allocation priority sorting in the data transmission scheduling configuration, extract the acquisition batches in the order of image frames with high bandwidth first, sequentially extract the single-frame image data within each acquisition batch, and then perform gray-scale processing on the image frames to reduce the impact of light changes on subsequent recognition. Extract the traffic facility image area by setting a gray-scale threshold. Assume that the gray-scale threshold is set to 120, and filter by comparing whether the pixel gray-scale value is greater than 120, and retain the areas that may be traffic facilities. At the same time, set the area threshold to 1000 pixel points to filter out small noise areas. For example, 3 traffic facility image blocks that meet the area requirements are extracted from the first frame of the image. Continue to extract the fixed sign image area in the image frame through contour detection. The contour detection uses the area with an edge intensity greater than the set boundary value of 200 as a candidate, and further filters the standard fixed sign templates through shape matching, such as circular speed limit signs and triangular warning signs. After identifying the traffic facility or fixed sign image blocks, extract the center point of the minimum circumscribed rectangle of each image block as the image center point, record the corresponding image frame number and the center point coordinate position, and organize and file the center point positions corresponding to all image frames to form an image center point coordinate set.

[0024] The trajectory change extraction sub-module calls the image center point coordinate set, extracts the horizontal displacement and vertical displacement of the center points of continuously acquired image frames, calculates the displacement increment data of adjacent frame center points, records the continuous displacement increment set, and generates the center point trajectory change amount; First, sequentially extract the horizontal coordinates of the center points of two adjacent frames according to the frame sequence and , the vertical coordinates and , respectively calculate the horizontal displacement increment and the vertical displacement increment . For example, the center point position of the first frame of the image is (100, 200), and the second frame is (105, 198), then the horizontal displacement increment , the vertical displacement increment . In this way, traverse all continuously acquired image frames to form a displacement increment set, and then record the horizontal and vertical increments between each pair of frames to form a continuous center point trajectory change sequence, and organize the sequence into a table structure for convenient subsequent camera offset analysis. For example, 9 groups of displacement increment data pairs are formed for 10 frames of images, and finally the center point trajectory change amount is generated.

[0025] Table 2 Data table of the center point of the image and the trajectory change: ; As shown in Table 2, the coordinates of the center point of the image frame and the displacement increment data of adjacent frames are listed. The data in this table is used for subsequent calculation of the camera offset level value.

[0026] The camera offset evaluation sub-module calls the amount of change in the center point trajectory, extracts the time series distribution of the continuous displacement increment set, and uses the formula: ; Calculate the camera offset level value, compare it with the preset camera offset level interval, and generate the camera offset value; Among them, represents the camera offset level value, represents the lateral displacement increment of the center point of the i-th frame, represents the longitudinal displacement increment of the center point of the i-th frame, means taking the maximum value, represents the number of consecutive sampling frames, represents the index number of the current sampling frame, represents the horizontal coordinate axis in the image coordinate system, represents the vertical coordinate axis in the image coordinate system; First, extract each group of lateral increments and longitudinal increments , organize them into a sequence according to the sampling time order, extract the increment data based on the time stamp order of each frame, and then use the formula: ; Calculate the camera offset level value. It is set that for the 9 groups of displacement data collected, , , substitute into the formula: ; An offset level value of approximately 6.946 is obtained. Subsequently, according to the set camera offset level interval, the offset level is compared with the offset level reference table. If the offset level is less than 5, it is determined as a slight offset; 5 to 10 is a medium offset; and greater than 10 is a severe offset. In this example, the offset level value of 6.946 belongs to the medium offset interval, and finally, the camera offset value is generated. Among them, the camera offset level value refers to a numerical index formed by comprehensively considering the overall average fluctuation level of the horizontal displacement increment and the vertical displacement increment in each frame and the single maximum offset amplitude during the continuous monitoring of the change of the center point of the image. This index is expressed in pixels and reflects the stability state of the camera within the set monitoring time window. The larger the value, the more obvious the offset phenomenon; the smaller the value, the higher the stability of the device. Its specific effect is to provide an objective quantitative basis for subsequent offset alarms, equipment inspections, and construction safety monitoring. It can accurately perceive the state change of the camera device through remote image data without relying on external sensors, improving the system's anomaly detection ability.

[0027] Please refer to Figure 4 , the path screening module includes: The trajectory point extraction sub-module calls the camera offset value, collects the monitoring image frame data, identifies the contours of the construction workers in the image, extracts the position coordinate points of the construction workers in each frame, arranges the position coordinate point sets in the order of time stamps, and generates a personnel trajectory data set; First, the construction area image frames are collected through the monitoring system at a frequency of 5 frames per second. In each frame of the image, the image is converted into a grayscale image using image processing methods, and a grayscale threshold of 120 is set to screen the contour boundary area. For all detected contour areas, small noise areas are further removed with a contour area threshold of 1000 pixel points, and only the contours with an area greater than 1000 pixel points are retained. Subsequently, the coordinates of the center point of the circumscribed rectangle of each contour area are extracted as the position points of the construction workers. For example, the coordinate point (120, 300) is extracted in the first frame, and the coordinate point (123, 305) is extracted in the second frame. Each extracted coordinate point records the corresponding timestamp information. For example, the timestamp of the first frame is 12:00:00.0, and the second frame is 12:00:00.2. Subsequently, all position coordinate points are arranged according to the timestamp order to ensure the consistency of position information under continuous time series. Finally, the position points of the construction workers corresponding to all timestamps form a trajectory point set, forming a continuous time-position mapping relationship, that is, a personnel trajectory data set.

[0028] The crossing event extraction sub-module calls the personnel trajectory data set, obtains the boundary data of the electronic fence in the work area, detects the spatial relationship status between each position coordinate point and the electronic fence boundary, detects the boundary crossing event, and records the trajectory points, crossing directions, and crossing time points where the boundary crossing occurs, generating a fence crossing event data set; Load the electronic fence boundary of the preset work area, which is represented by polygon coordinates. Set four corner points to form a rectangular fence range. For example, the fence corner point coordinates are (50, 50), (50, 500), (500, 500), (500, 50). Judge the spatial relationship between all construction worker trajectory points and the fence boundary. Determine the internal state by comparing whether the horizontal and vertical coordinates of the trajectory points fall within the fence coordinate range. If the horizontal or vertical coordinate of the trajectory point exceeds the corresponding fence range, it is determined as a fence crossing event. For each crossing event, record the trajectory point coordinates at the time of crossing, such as (510, 300), record the crossing direction. Set the direction to be judged by the included angle relationship between the direction vectors before and after the movement of the trajectory point and the fence boundary vector. For example, the out-of-bounds direction is towards the positive X-axis on the right. The crossing time point takes the corresponding timestamp information, such as 12:00:01.2. Finally, organize and collect the trajectory points, crossing directions, and crossing time points of all crossing events to generate a fence crossing event dataset.

[0029] The out-of-bounds trajectory screening sub-module calls the fence crossing event dataset, compares it with the crossing direction and the data of the external boundary direction of the work area, extracts the trajectory segments whose crossing directions point to non-work areas, and generates out-of-bounds behavior recognition records; Extract each crossing record in the fence crossing event dataset, read the position of the crossing trajectory point and the crossing direction parameter, and then call the data of the external boundary of the work area. The external boundary direction is defined by the standard due east, due south, due west, and due north directions. Calculate the included angle between each crossing direction vector and the external boundary direction. Set the included angle less than 30 degrees as the same-direction judgment condition, that is, if the included angle between the crossing direction and the external boundary direction of the work area is less than 30 degrees, then determine that this trajectory segment is a trajectory pointing to a non-work area. For example, if the crossing direction is due east and the east side of the work area boundary is set as a non-work area, then this crossing is confirmed as an out-of-bounds crossing. Extract the corresponding crossing trajectory segment, including the starting point, ending point, and intermediate trajectory data of the crossing, to form a complete trajectory subsequence. Finally, the set of all trajectory segments that meet the out-of-bounds judgment criteria is generated as an out-of-bounds behavior recognition record.

[0030] Table 3 Data table of personnel trajectory and fence crossing: ; As shown in Table 3, it lists the position coordinates, spatial relationship status, whether to cross the fence, and the corresponding crossing direction of construction workers at each time point. The data in the table is used for subsequent out-of-bounds trajectory screening.

[0031] Please refer to Figure 5 , the area monitoring module includes: The position trajectory extraction sub-module obtains the out-of-bounds behavior recognition record, identifies the dangerous area signs in the highway construction area, sets the spatial boundary of the dangerous area, and obtains the dangerous area boundary dataset; First, all image frames of the construction activity areas are extracted based on the personnel trajectory data contained in the out-of-bounds behavior recognition records. Subsequently, the image analysis module is called to perform static object extraction on each frame of the image, extracting fixed facilities and temporary construction area identifiers in the image, and screening them using color features and shape templates. For example, a red rectangular area is set as the dangerous area identifier. After identifying the areas that meet the features, the corresponding area boundary contour coordinates are extracted. For example, for a certain dangerous area, the four corner coordinates are identified as (100, 200), (100, 400), (300, 400), and (300, 200). Further, a spatial boundary dataset is established based on the boundary coordinates identified for each dangerous area. The spatial boundary dataset is stored in the form of a closed polygon boundary for each dangerous area. Finally, by integrating the coordinate information of each dangerous area, a dangerous area boundary dataset is generated.

[0032] The dangerous proximity screening sub-module calls the dangerous area boundary dataset, extracts the sequence of distance values and the corresponding time series between the current position coordinate points of the personnel and the dangerous area boundary points, identifies the continuous time intervals and position change amounts, and uses the formula: ; Calculate the proximity speed risk score, and obtain the dangerous proximity behavior detection record by comparing it with the preset dangerous area rapid approach risk threshold; Among them, represents the proximity speed risk score, represents the shortest distance between the construction personnel and the dangerous area boundary at the k-th moment, represents the time stamp at the k-th moment, represents the preset standard approach speed of the system, represents a very small positive number to prevent division by zero, represents the shortest distance value between the construction personnel and the dangerous area boundary at the (k + 1)-th moment, represents the time stamp at the (k + 1)-th moment, represents the current sampling segment index number, represents the total number of time periods included in the continuous trajectory sequence; First, extract the current position horizontal and vertical coordinates based on the trajectory points of the construction personnel, and call the dangerous area boundary dataset to calculate the shortest straight-line distance from the current point of the construction personnel to the nearest boundary segment of each dangerous area boundary. Set the unit time interval to 0.5 seconds, and record the shortest distance value and the time stamp , for example, the first time point is 12:00:00.0, , the second time point is 12:00:00.5, , and then use the formula: ; Calculate the approach speed risk score, where The value is set to 0.01 to prevent division by zero, The standard approach speed is set to 1.5 meters per second. Substitute the data example, with three consecutive groups of distances , , , and the time interval is 0.5 seconds each. Calculate: ; ; Substitute into the formula and expand the calculation: ; Finally, obtain the approach speed risk score , and according to the set dangerous approach risk threshold of 1.0, if the risk score is greater than 1.0, it is determined as a dangerous approach behavior. This example belongs to a dangerous approach behavior, and finally a dangerous approach behavior detection record is formed. Among them, the approach speed risk score refers to a dimensionless risk quantification index formed by comprehensively considering the speed change rate of approaching the dangerous area, the dynamic approach distance weight, and the standardized speed benchmark during the continuous monitoring of the position trajectory of construction workers. The index has no unit, and the higher the value, the more prominent the risk behavior of construction workers approaching the dangerous area at a faster speed and shorter distance in a short period of time. The lower the value, the milder the approaching trend and the lower the risk. It aims to provide a real-time quantitative evaluation basis for the dynamic behavior of construction workers approaching the dangerous area within a continuous time period for the remote intelligent monitoring system, and support dangerous behavior determination, alarm triggering, and safety management decision-making.

[0033] The unprotected crossing recognition sub-module calls the dangerous approach behavior detection record, real-time detects the wearing status of the personal protective equipment of the construction workers, identifies and extracts the dangerous area crossing behavior in the unprotected state, and generates a safety behavior recognition record; First, the abnormal trajectory segments in the dangerous approach behavior detection records are extracted, and the image frame data corresponding to the trajectory segments are extracted. The protective equipment status is identified for each frame of the image. The identification objects are set as reflective vests and safety helmets. The construction personnel area is extracted in the image frame, and the reflective vest area is extracted by color features. The width ratio of the yellow reflective tape is used to determine whether the reflective vest is worn. The reflective area ratio is set to be greater than 10% to be considered as wearing. Safety helmet detection is based on the extraction of color features in the head area. Blue or yellow block areas are detected. If the area exceeds 20% of the head area, it is considered to be wearing a safety helmet. If a reflective vest is detected in the first frame but a safety helmet is not detected, and no protective equipment is detected in the second frame, then the unprotected state is recorded. Combined with the dangerous approach trajectory segment, the trajectory of the unprotected person is compared with the boundary of the dangerous area. When the person's position point crosses the boundary of the dangerous area, a crossing event is recorded. Finally, all unprotected crossing behaviors are collected to generate a safe behavior recognition record.

[0034] Table 4 Hazardous area approach and unprotected detection data table: ; As shown in Table 4, the changes in the shortest distance, approach speed, risk score and protection status determination results of construction workers around the dangerous area are listed. The data in the table are used to screen unprotected crossing behaviors.

[0035] See also Figure 6 , the low light monitoring module includes: The reflective trajectory extraction submodule calls the safety behavior recognition record, obtains image data in a low-light environment at night and identifies reflective signs in the image, extracts the position coordinates of reflective bright spots in continuous frame images, records the reflective bright spot trajectory path in a continuous time series, and generates a reflective trajectory path dataset; Firstly, all the construction workers’ trajectory segments are extracted from the safety behavior recognition records, and the night monitoring image frame data of the corresponding time period is called to pre-process the image. The brightness is adjusted to 0.7 times the ambient brightness standard set by the standardized level. The brightness threshold is set to 200 through threshold segmentation to filter the image highlight area, identify potential reflective signs, and extract the centroid coordinates of the area with the maximum brightness in each frame of the image as the reflective bright spot position. For example, the bright spot coordinates of the first frame are (300,500), and the bright spot coordinates of the second frame are (302,505). The corresponding timestamp sequence is recorded with a sampling frequency of 5 frames per second. The continuous bright spot coordinates are arranged according to the timestamp to form a continuous reflective bright spot trajectory path. Each reflective bright spot trajectory is encapsulated into a data structure for storage, and finally a reflective trajectory path dataset is generated.

[0036] The trajectory anomaly identification submodule calls the reflective trajectory path dataset, analyzes the continuity of the change direction between trajectory points and the trajectory interruption frequency, detects abnormal trajectory segments, and generates an abnormal trajectory segment set; First, for each reflective bright spot trajectory, a trajectory point sequence is extracted, and the motion direction vector is calculated based on two adjacent trajectory points. The direction vector angle change threshold is set to 30 degrees, and the direction change of each pair of adjacent trajectory segments is extracted. If the direction change angle is greater than 30 degrees, it is marked as an abnormal inflection point. For example, the direction change angle from the 3rd point to the 4th point is 45 degrees, which is determined to be an abnormal inflection. The number of abnormal inflection points is accumulated, and the abnormal inflection ratio in the trajectory segment is set to exceed 10%, which is used to determine that the trajectory direction continuity is abnormal. At the same time, if the reflective bright spot is not detected for more than a set time interval of 3 frames, it is determined to be a trajectory interruption. For example, if 3 frames are continuously missing from 12:00:01.0 to 12:00:01.6, it is determined to be an interruption. All abnormal trajectory segments are extracted, and the starting and ending points of the abnormal trajectory are marked. All detected abnormal trajectory segments are sorted into a set, and finally an abnormal trajectory segment set is generated.

[0037] The abnormal trajectory spatial detection submodule calls the abnormal trajectory segment set, combines the dangerous area boundary data, detects the spatial interaction relationship between the abnormal trajectory segment and the dangerous area boundary, detects abnormal movement trajectories at night, and generates nighttime behavior monitoring records; The position coordinates of the starting point and the ending point of each abnormal trajectory segment are extracted, and the boundary rectangle range of each dangerous area in the dangerous area boundary data set is called to perform spatial relationship judgment. If the coordinates of the starting point or the ending point of the abnormal trajectory segment fall within the boundary rectangle of the dangerous area, it is determined that the trajectory segment has a spatial interaction relationship with the dangerous area. For example, if the starting point (310, 510) of the trajectory segment falls within the range of (300, 500) to (350, 550) of the dangerous area boundary, the trajectory segment is determined to be a dangerous interaction trajectory. The trajectory segment number, interactive dangerous area number, and interaction time are recorded. Finally, all abnormal trajectory segments with spatial interaction relationships are collected to generate nighttime behavior monitoring records.

[0038] Table 5 Reflection trajectory anomaly detection and spatial interaction data table: ; As shown in Table 5, the trajectory data of reflective bright spots at night, the direction change angle, the interruption status and the interactive judgment of the danger zone are listed. The data in the table are used for abnormal trajectory space detection and nighttime behavior monitoring.

[0039] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent remote monitoring and identification system for highway construction safety behavior, characterized in that: The system comprises: The scheduling management module obtains the monitoring image data frame, extracts the timestamp, picture clarity level, and regional priority tag of each frame, evaluates the urgency and importance of the image frame, adjusts the bandwidth resource allocation parameters of each image frame based on the available bandwidth level of the node, and generates the data transmission scheduling configuration; The center detection module calls the data transmission scheduling configuration, analyzes the image frame and identifies static traffic facilities and fixed signs, obtains the center point of the image, identifies the change of the center point trajectory of consecutive frames, identifies the offset risk of the camera angle, and generates the camera offset value; The path screening module calls the camera offset value, identifies the movement trajectory of personnel by analyzing the surveillance image, and records the number of fence crossings, crossing directions and crossing time points in combination with the electronic fence boundary data, extracts the crossing trajectory pointing to the non-operating area, and generates a cross-border behavior identification record; The area monitoring module obtains the continuous position information of the personnel based on the cross-border behavior identification record, extracts the distance change sequence between the current position and the boundary of the danger zone, combines the identification result of the wearing status of the protective equipment, detects the dangerous zone crossing behavior under unprotected conditions, and generates a safe behavior identification record.

2. The intelligent remote monitoring and identification system for highway construction safety behavior according to claim 1 is characterized in that: The data transmission scheduling configuration includes image frame transmission priority sorting information, bandwidth resource allocation ratio data, and node signal load status records. The camera offset value specifically includes the center point trajectory offset amplitude, the center point offset direction change, and the camera posture offset angle. The cross-border behavior identification record includes the fence crossing times statistics, the non-operating area pointing trajectory sequence, and the crossing event timestamp list. The safety behavior identification record specifically includes the fast crossing action judgment result, the unprotected state identification label, and the danger zone approach speed change curve.

3. The intelligent remote monitoring and identification system for highway construction safety behavior according to claim 1 is characterized in that: The scheduling management module includes: The image attribute extraction submodule obtains the monitoring image data frame, extracts the timestamp parameter, picture clarity level, and regional priority label of each frame, and generates image attribute information; The image frame priority evaluation submodule calls the image attribute information, evaluates the urgency value and importance level value of each image frame according to the regional priority label and the picture clarity level, and generates an image frame importance evaluation index; The bandwidth resource dynamic configuration submodule calls the image frame importance evaluation index, combines the node available bandwidth level parameter, extracts the node data transmission demand parameter, the node remaining bandwidth capacity parameter and the node load distribution state parameter, calculates the image frame bandwidth allocation adjustment amount, and obtains the data transmission scheduling configuration.

4. The intelligent remote monitoring and identification system for highway construction safety behavior according to claim 3 is characterized in that: The center detection module includes: The image center extraction submodule obtains the data transmission scheduling configuration, collects monitoring image frame data, obtains the center point in each frame of the image by analyzing the image frame and identifying the traffic facility image data and the fixed sign image data, and generates an image center point coordinate set; The trajectory change extraction submodule calls the image center point coordinate set, extracts the lateral displacement and longitudinal displacement of the center point of the continuously acquired image frames, calculates the center point displacement increment data of the adjacent frames, records the continuous displacement increment set, and generates the center point trajectory change; The camera offset assessment submodule calls the center point trajectory change, extracts the time series distribution of the continuous displacement increment set, calculates the camera offset level value, and compares it with the preset camera offset level interval to generate a camera offset value.

5. The intelligent remote monitoring and identification system for highway construction safety behavior according to claim 4 is characterized in that: The path screening module includes: The trajectory point extraction submodule calls the camera offset value, collects monitoring image frame data, identifies the outline of the construction personnel in the image, extracts the position coordinate points of the construction personnel in each frame, arranges the position coordinate point set in sequence according to the timestamp, and generates a personnel trajectory data set; The crossing event extraction submodule calls the personnel trajectory dataset, obtains the electronic fence boundary data of the work area, detects the spatial relationship state between each position coordinate point and the electronic fence boundary, detects the boundary crossing event, and records the trajectory point, crossing direction and crossing time point where the boundary crossing occurs, and generates a fence crossing event dataset; The boundary crossing trajectory screening submodule calls the fence crossing event data set, compares the crossing direction with the boundary direction data of the outer limit of the working area, extracts the trajectory segment whose crossing direction points to the non-working area, and generates a boundary crossing behavior recognition record.

6. The intelligent remote monitoring and identification system for highway construction safety behavior according to claim 5 is characterized in that: The area monitoring module includes: The position trajectory extraction submodule obtains the cross-border behavior recognition record, identifies the dangerous area mark in the highway construction area, sets the spatial boundary of the dangerous area, and obtains the dangerous area boundary data set; The dangerous approach screening submodule calls the dangerous area boundary data set, extracts the distance value sequence and corresponding time series between the current position coordinate point of the person and the dangerous area boundary point, identifies the continuous time interval and the position change, calculates the approach speed risk score, and obtains the dangerous approach behavior detection record by comparing it with the preset dangerous area fast approach risk threshold; The unprotected crossing identification submodule calls the dangerous approach behavior detection record, detects the wearing status of protective equipment of construction personnel in real time, identifies and extracts the dangerous area crossing behavior in the unprotected state, and generates a safe behavior identification record.

7. The intelligent remote monitoring and identification system for highway construction safety behavior according to claim 1 is characterized in that: The system further comprises: The low-light monitoring module calls the safety behavior recognition record, obtains image data in a low-light environment at night and identifies reflective marks in the image, extracts the trajectory path of reflective bright spots in continuous frames, analyzes the trajectory direction continuity and interruption frequency, identifies abnormal trajectory segments and compares the spatial relationship with the boundary of the dangerous area, detects abnormal movement trajectories at night, and generates night behavior monitoring records; The nighttime behavior monitoring records specifically refer to the nighttime trajectory continuity score, abnormal trajectory offset distance, and abnormal trajectory distribution range in dangerous areas.

8. The intelligent remote monitoring and identification system for highway construction safety behavior according to claim 7 is characterized in that: The low light monitoring module comprises: The reflective trajectory extraction submodule calls the safety behavior recognition record, obtains image data in a low-light environment at night and identifies reflective marks in the image, extracts the position coordinates of reflective bright spots in continuous frame images, records the reflective bright spot trajectory path in a continuous time series, and generates a reflective trajectory path data set; The track anomaly identification submodule calls the reflective track path data set, analyzes the continuity of the change direction between track points and the track interruption frequency, detects abnormal track segments, and generates an abnormal track segment set; The abnormal trajectory spatial detection submodule calls the abnormal trajectory segment set, combines the dangerous area boundary data, detects the spatial interaction relationship between the abnormal trajectory segment and the dangerous area boundary, detects abnormal movement trajectories at night, and generates nighttime behavior monitoring records.

Images