5g-based road construction video inspection and data analysis system

Abstract

The present application relates to the technical field of video monitoring, in particular to a road construction video inspection and data analysis system based on 5G, which comprises a video frame sequence extraction module, a work behavior fluctuation identification module, a frequent segment aggregation positioning module, a channel segment frequency hopping screening module and a video paragraph scheduling triggering module.In the present application, the action fluctuation frame segment is positioned based on the target edge displacement and the image overlap change, the behavior characteristic segment is constructed through high-frequency area clustering, the continuous camera sequence is extracted in combination with the camera direction and the number connection relationship, the channel stable segment is screened according to the frequency hopping change difference value, the scheduling mapping is established by fusing the time stamp and the task identification, the whole process processing from behavior detection, area aggregation, camera selection to channel stability discrimination is realized, and the accuracy of key behavior extraction, the continuity of camera connection and the stability of video scheduling are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of video monitoring, and in particular to a road construction video inspection and data analysis system based on 5G. BACKGROUND

[0002] The technical field of video monitoring belongs to an important branch of information and communication technology, and its core matters include image acquisition, video coding, transmission control, target recognition, behavior analysis, and remote management, etc. Its application scenarios cover urban security, traffic supervision, industrial inspection, public facility management, and other aspects. With the development of 5G communication technology, real-time transmission of high-definition video and intelligent analysis have become the key direction of the evolution of video monitoring technology, which changes it from passive recording to active sensing and intelligent early warning system, forming a new generation of video monitoring system characterized by data-driven and intelligent linkage. Among them, the traditional road construction video inspection refers to a way of relying on manual patrol or laying fixed cameras to collect video information, which is used to monitor the construction site conditions and judge the operation safety. This way mainly uses manual observation of video pictures and offline recording for analysis and judgment, which cannot realize real-time aggregation and analysis of multi-source high-definition video, nor can it realize low-latency transmission and remote scheduling management of video data based on 5G communication network.

[0003] The prior art lacks fine-grained recognition means for continuous changes of image target behavior in the video content processing process, and the fragment recognition relies on single-frame observation, which leads to time jump problem in motion analysis. When the operation area overlaps or the scene changes complexly, it cannot realize effective aggregation of image area. The lens processing does not distinguish the direction attribute and the numbering connection relationship, causing the lens extraction to be fragmented. The video fragment scheduling does not consider channel changes, leading to the decline of fragment effectiveness in high interference scenes. The data processing logic mainly uses static frame acquisition and coarse-grained judgment, and lacks joint analysis capability of image behavior dimension and signal channel dimension, which leads to untimely capture of key operation behavior and lack of pertinence of lens content scheduling. SUMMARY

[0004] The purpose of the present application is to solve the shortcomings in the prior art, and a road construction video inspection and data analysis system based on 5G is proposed.

[0005] In order to achieve the above purpose, the present application adopts the following technical scheme: the road construction video inspection and data analysis system based on 5G comprises:

[0006] The video frame sequence extraction module acquires 5G videos of road paving sections and intersection construction sections, extracts the spatial profile of the target in the continuous frames, calculates the edge displacement and overlap ratio, judges whether it exceeds the offset threshold, and outputs the action fluctuation frame time list;

[0007] The operation behavior fluctuation identification module locks a video segment according to the action fluctuation frame time list, counts the image region number with the highest occurrence frequency, clusters and groups if there is no shot switching mark, and obtains an image segment number set;

[0008] The frequent segment aggregation positioning module calls the image segment number set, reads the shot direction and time label, calculates the number continuity and direction consistency proportion, and outputs a continuous shot segment sequence;

[0009] The channel segment frequency hopping screening module reads the continuous shot segment sequence, obtains the 5G frequency point change of the corresponding video segment, calculates the frequency point difference value and compares the frequency hopping smooth value, and outputs a signal stable video segment number set;

[0010] The video paragraph scheduling trigger module calls the signal stable video segment number set, extracts the timestamp and task field, establishes a scheduling index combined with the task order value, records as a shot mapping item, and outputs a road construction video inspection and data analysis result.

[0011] As a further scheme of the application, the action fluctuation frame time list includes frame segment start and end time indexes, target edge displacement values, and image overlap proportion values, the image segment number set includes occurrence region numbers, region cluster numbers, and shot switching marks, the continuous shot segment sequence includes shot number sequences, shot direction consistency, and number continuity indicators, the signal stable video segment number set includes frequency point variation ranges, frequency hopping stability thresholds, and stable segment numbers, and the road construction video inspection and data analysis result includes shot start and end timestamps, task identification codes, video channel information, and scheduling order indexes.

[0012] As a further scheme of the application, the video frame sequence extraction module includes:

[0013] The video data receiving submodule obtains the image frame sequence collected by the 5G high-definition video monitoring points deployed in the road paving section and the intersection construction section, detects the position of the target object in the continuous frame, extracts the spatial contour boundary of the target in each frame, and generates a target frame contour region set;

[0014] The contour edge calculation submodule compares the target edge coordinates in adjacent frames based on the target frame contour region set, calculates the average displacement distance value and the contour overlap proportion value, compares the displacement distance value with the frame edge offset cumulative threshold value, obtains the frame index number greater than the frame edge offset cumulative threshold value, and obtains an image frame displacement offset index list;

[0015] The frame sequence number generation submodule calls the image frame displacement offset index list, screens the frame index numbers of the continuous change section, matches the set time window range, establishes a time axis mapping relationship, and generates an action fluctuation frame time list.

[0016] As a further scheme of the present application, the operation behavior fluctuation identification module comprises:

[0017] The video segment positioning submodule matches the corresponding frame image interval in the original video sequence according to the frame segment start and end time indexes in the motion fluctuation frame time list, intercepts the continuous image frame sequence in the frame segment interval, records the frame index range of the sequence in the video, and generates a fluctuation frame segment image sequence set;

[0018] The region number statistics submodule calls the fluctuation frame segment image sequence set, detects the region number corresponding to each frame in the image region, counts the cumulative occurrence number of the number in the frame segment, extracts the region number with the maximum number of occurrences, and judges whether there is a shot switching mark according to the change order of the region number in the continuous frames, and generates a region number stability state;

[0019] The segment number grouping submodule filters out the frame segment numbers with the shot switching mark according to the region number stability state, calls the region numbers counted in the frame segment without shot switching, clusters and merges according to the number fitting degree and spatial position adjacency relationship, and obtains an image segment number set.

[0020] As a further scheme of the present application, the frequent segment aggregation positioning module comprises:

[0021] The numbered segment reading submodule calls the segments corresponding to the numbers in the image segment number set, positions the segment frame segment in the video data, extracts the shot direction attribute and start and end time label in the frame segment, establishes a mapping index of the number and direction and time, and generates a number attribute mapping table;

[0022] The shot attribute extraction submodule extracts the shot direction value and time label in the number group according to the number attribute mapping table, arranges the number frame segment in time order, extracts the shot direction sequence, and obtains a number shot direction sequence;

[0023] The continuity matching identification submodule detects whether the number order is continuous according to the number shot direction sequence, counts the number of numbers with consistent direction values, calculates the number continuity consistency value, judges whether it is higher than the set identification standard threshold, selects the number sequence that meets the condition, and generates a continuous shot segment sequence.

[0024] As a further scheme of the present application, the channel segment frequency hopping screening module comprises:

[0025] The shot segment extraction submodule extracts the video segment corresponding to the number based on the number listed in the continuous shot segment sequence, analyzes the frequency point sequence information recorded in the road side edge 5G signal tracking node, and arranges it into a sequence set in the number order, and obtains a video frequency point sequence set;

[0026] The frequency difference calculation submodule calls the video frequency sequence set, obtains the start and end frequency values ​​of the video segment for the frequency hopping frequency values ​​recorded at the start and end points of each video segment, calculates the difference range and records the difference data interval in sequence, calculates the frequency hopping difference metric value of the video segment, and obtains the frequency hopping change amplitude set by combining the difference change amplitude and the frequency dispersion.

[0027] The frequency hopping stability filtering submodule compares the frequency hopping difference metric of the video segment with the set signal frequency hopping stability threshold one by one based on the frequency hopping variation amplitude set, and filters out all video segment numbers whose frequency hopping difference metric is less than the signal frequency hopping stability threshold to obtain a set of signal stable video segment numbers.

[0028] As a further aspect of the present invention, the video segment scheduling triggering module includes:

[0029] The timestamp extraction submodule calls the segment numbers in the signal-stable video segment number set, reads the first frame timestamp and the last frame timestamp in the video segment corresponding to each number, and performs number mapping processing on the time information to generate a video segment time index table.

[0030] The task information binding submodule extracts the video task identifier field and video channel binding information for each time interval according to the video segment time index table, reads the task sequence value in chronological order, and establishes a binding record index to obtain the task channel binding structure.

[0031] The scheduling mapping generation submodule matches the start and end boundaries of adjacent task number combinations according to the task sequence value, first and last frame timestamps and channel information in the task channel binding structure, sets the scheduling start and end numbers, establishes a bidirectional mapping relationship between the shot number and the task scheduling sequence, generates scheduling task shot mapping items, and obtains the road construction video inspection and data analysis results.

[0032] As a further aspect of the present invention, the process of obtaining the continuity consistency value of the number is to determine whether the direction values ​​of two adjacent lenses in the numbered lens direction sequence are consistent, and to statistically accumulate the number of consistency values ​​using a set weighting ratio.

[0033] In the process of determining whether the value is higher than the set recognition standard threshold, the judgment is made based on the ratio between the number of times the direction value is consistent and the total number of the numbered lens direction sequence, combined with the preset continuity evaluation rules.

[0034] As a further aspect of the present invention, in the process of obtaining the video frequency point sequence set, each video segment corresponding to the number is limited to include a start frequency point value and an end frequency point value, and both of them have a time tag;

[0035] In the process of comparing the frequency hopping difference metric of the video segment with the set signal frequency hopping stability threshold one by one, the signal frequency hopping stability threshold is determined based on the dispersion of the frequency hopping change amplitude set, and the set of signal stable video segment numbers is verified to be continuously arranged on the time label.

[0036] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0037] In this invention, action fluctuation frames are located based on target edge displacement and image overlap changes. Behavioral feature segments are constructed through high-frequency region clustering. Continuous shot sequences are extracted by combining shot direction and numbering connection relationships. Channel-stable segments are selected based on frequency hopping change differences. A scheduling mapping is established by fusing timestamps and task identifiers. This achieves full-process processing from behavior detection, region aggregation, shot selection to channel stability judgment, improving the accuracy of key behavior extraction, the continuity of shot connection, and the stability of video scheduling. Attached Figure Description

[0038] Figure 1 This is a system flowchart of the present invention;

[0039] Figure 2 This is a flowchart of the video frame sequence extraction module of the present invention;

[0040] Figure 3 This is a flowchart of the work behavior fluctuation identification module of the present invention;

[0041] Figure 4 This is a flowchart of the frequent fragment aggregation and positioning module of the present invention;

[0042] Figure 5 This is a flowchart of the channel segment frequency hopping screening module of the present invention;

[0043] Figure 6 This is a flowchart of the video segment scheduling trigger module of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the 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 merely illustrative and not intended to limit the invention.

[0045] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0046] Please see Figure 1 The 5G-based road construction video inspection and data analysis system includes:

[0047] The video frame sequence extraction module acquires the content collected by 5G high-definition video monitoring points deployed in the road paving section and the intersection construction section, extracts the spatial contour position of the same target object in consecutive image frames, calculates the edge displacement distance of the target area and the image overlap ratio in each pair of frames, determines whether the cumulative displacement value in consecutive frames exceeds the cumulative displacement threshold and generates a set of frame sequence numbers to obtain a list of motion fluctuation frame times.

[0048] The operation behavior fluctuation recognition module locks the corresponding segment in the video according to the start and end time index of the frame segment in the action fluctuation frame time list, counts the region number that appears most frequently in the image region in the segment, and determines whether there is a shot switching mark in the adjacent region numbers. If there is no shot switching, the region numbers are clustered and grouped to obtain the image segment number set.

[0049] The frequent segment aggregation and localization module calls the corresponding segments in the image segment number set, reads the lens direction attribute and start and end time tags in the segment, calculates the number continuity and direction matching ratio of each group of segments, and judges whether the ratio reaches the set recognition standard, filters out the lens number sequence, and obtains the sequence of continuous lens segments.

[0050] The channel segment frequency hopping screening module reads the video segments corresponding to the numbers listed in the sequence of continuous shot segments, obtains the frequency hopping frequency point change value recorded in the 5G signal tracking node on the roadside edge, calculates the frequency point change difference range between the start and end points of the segment and compares it with the signal frequency hopping stability value, sets and filters out the video segment numbers that meet the frequency hopping stability standard, and obtains the signal stable video segment number set.

[0051] The video segment scheduling trigger module calls the segment number in the set of stable video segment numbers, reads the first frame timestamp and last frame timestamp corresponding to each number, extracts the task identifier field and video channel binding information within the time period, and establishes a shot scheduling start and end index in combination with the task sequence value, which is recorded as the shot mapping item of the scheduling task, and obtains the results of road construction video inspection and data analysis.

[0052] The motion fluctuation frame time list includes frame start and end time indexes, target edge displacement values, and image overlap ratio values. The image segment number set includes occurrence area numbers, area cluster numbers, and shot switching identifiers. The continuous shot segment sequence list includes shot number sequence, shot direction consistency, and number continuity index. The signal stable video segment number set includes frequency point variation range, frequency hopping stability threshold, and stable segment number. The road construction video inspection and data analysis results include shot start and end timestamps, task identifier codes, video channel information, and scheduling order indexes.

[0053] Please see Figure 2 The video frame sequence extraction module includes:

[0054] The video data receiving submodule acquires the image frame sequence collected by 5G high-definition video monitoring points deployed in the road paving section and the intersection construction section, detects the position of the target object in the continuous frame, extracts the spatial contour boundary of the target in each frame, and generates a set of target frame contour regions.

[0055] The video data receiving submodule acquires video image streams from 5G high-definition video monitoring points deployed at road paving sections and intersection construction sections. These monitoring points acquire 30 frames of high-definition images every second, transmitted in 1080p format. The receiving submodule first decodes the video stream, extracts image data frame by frame, and converts it into grayscale images. Then, it processes pixel changes between consecutive frames to extract areas that may contain target objects. During execution, the pixel grayscale change threshold Δp is initially set to 20. This value is derived from on-site test data analysis. Statistical analysis of the grayscale difference between the background and targets (such as construction vehicles, paving machinery, and workers) under various lighting and construction backgrounds revealed that the pixel difference between the target area and the background is generally between 18 and 45. Therefore, setting the threshold to 20 effectively identifies most target areas while avoiding misidentification caused by slight background fluctuations. The submodule calculates the grayscale difference of pixels at corresponding positions in the current frame and the previous frame pixel by pixel. For example, in frame F001, a pixel has a grayscale value of 120 in frame t-1 and 143 in frame t, with a grayscale difference of 23. If the value is greater than 20, the pixel position is recorded as a change point. After multiple change points form a candidate region, the boundary contour extraction operation is then performed. Pixel connectivity analysis is used to extract closed boundaries and form a contour region set. In this process, to avoid extracting too small areas such as vehicle reflections, dust, and other noise, the contour area threshold A_min is set to 300 pixels. This value is obtained by statistically analyzing the contour areas of typical targets (such as tires and human heads) in 1000 frames of images. The smallest effective target area is 326 pixels, and 90% of the target area is concentrated between 400 and 800 pixels. Therefore, 300 pixels is taken as the lower limit of the filter to further filter areas smaller than this area. In the example, 12 contours are extracted in frame F001. Among them, 3 contours with an area less than 300 pixels are removed, and 9 are retained for subsequent processing. Before removal, the contours, such as region A, contain 18 pixels forming an approximate rectangle with a width of 6 pixels and a height of 5 pixels, and a total area of ​​6 × 5 = 30 pixels, which is far below the set threshold. Therefore, it is judged as a noise region, and the target frame contour region set is generated.

[0056] The contour edge calculation submodule is based on the target frame contour region set, compares the target edge coordinates in adjacent frames, calculates the average displacement distance value and the contour overlap ratio value, compares the displacement distance value with the cumulative frame edge offset threshold, obtains the frame index number that is greater than the cumulative frame edge offset threshold, and obtains the image frame displacement offset index list.

[0057] The contour edge calculation submodule reads the target contour region set. Multiple target contour regions are recorded in each frame. This submodule extracts two consecutive frames from the cache each time, such as frames F001 and F002, compares the contour boundary points of the same target region, and extracts five corner points on the boundary of each contour as a keypoint set. The spatial displacement values ​​of these keypoints in adjacent frames are evaluated. The displacement values ​​are obtained by averaging the position differences of the same contour points in adjacent frames. To set a reasonable threshold, the team statistically analyzed the boundary point displacements of targets under normal movement conditions in different construction environments. They found that the target boundary displacement between most frames was between 2 and 9 pixels, while when the target undergoes a sudden movement (such as sudden start-up or rapid turning), its displacement usually exceeds 10 pixels. Therefore, a displacement threshold of 10 pixels was set. If the average keypoint displacement of any target contour exceeds this value, its frame number is recorded, such as between frames F002 and F003. The displacements of a target boundary point set are {9, 12, 11, 10, 13} pixels, with an average of 11 pixels, exceeding a set threshold. Therefore, frame F003 is recorded in the offset frame list. In addition, the overlap ratio of the contour region between frames is evaluated. For this purpose, the threshold R_min is defined as 0.6. This threshold comes from the analysis of the inter-frame overlap during the continuous movement of the target in the experiment. Statistical analysis shows that if the overlap of the same target in two frames is less than 60%, it usually indicates that it has been significantly displaced or has undergone occlusion. Therefore, the lower limit of the overlap ratio is set to 0.6. In the example, if the overlap area between frames F002 and F003 is 280 pixels, the total merged area is 530 pixels, and the overlap ratio is 280 / 530≈0.528, which is lower than the set threshold. Similarly, frame F003 is judged to meet the offset frame condition and recorded as an index. Finally, all frames that meet the condition of key point displacement greater than 10 pixels and overlap ratio less than 0.6 are used to obtain the image frame displacement offset index list.

[0058] The frame sequence number generation submodule calls the image frame displacement offset index list, filters the frame index numbers of continuously changing segments, matches the set time window range, establishes the time axis mapping relationship, and generates a list of motion fluctuation frame times.

[0059] After obtaining the image frame displacement offset index list, the frame sequence number generation submodule processes the frame numbers. First, it sorts the frames in ascending order according to their numbers. Then, it checks whether the interval difference Δf between the indices is 1 to identify whether the frames form a continuous action change segment. For example, if the offset frame list is {4, 5, 6, 9, 10, 11, 12, 17}, the system compares the number differences between adjacent frames and finds that {4, 5, 6} and {9, 10, 11, 12} form a continuous segment, while 17 is an independent frame. Next, it needs to evaluate whether each segment meets the minimum time window requirement. Here, the time window W is set to 1 second, and the image frame rate is 30fps. Therefore, a continuous frame count ≥ 30 frames is required to be considered a valid change segment. This threshold is derived from the analysis of the shortest duration of actions during road paving. Actual measurements show that most key operations, such as paver movement and roller start-up... The duration of actions such as movements must be at least 1 second. Therefore, 1 second is selected as the minimum action duration window. If a consecutive frame segment has less than 30 frames, it is discarded. This segment is considered an invalid short-term jitter. For example, the segment {4, 5, 6} has only 3 frames, which does not meet the condition and is discarded. However, if another segment, such as {30~60}, has 31 frames, which exceeds 30 frames, it is retained and converted into a timeline format. That is, the starting frame 30 corresponds to the time 30 / 30=1.0 seconds, and the ending frame 60 corresponds to the time 60 / 30=2.0 seconds. The action fluctuation time period is recorded as [1.0s–2.0s]. Repeating this process can convert multiple segments that meet the conditions into a time period list, such as {[1.0s–2.0s], [3.2s–4.5s], [5.0s–6.2s]}, generating an action fluctuation frame time list for downstream modules to perform tasks such as behavior pattern recognition or key operation statistics.

[0060] Please see Figure 3 The operation behavior fluctuation identification module includes:

[0061] The video segment localization submodule matches the corresponding frame image interval in the original video sequence based on the start and end time index of the frame segment in the motion fluctuation frame time list, extracts the continuous image frame sequence within the frame segment interval, records the frame index range of the sequence in the video, and generates a set of fluctuation frame segment image sequences.

[0062] After receiving the motion fluctuation frame time list, the video segment localization submodule first reads the start and end times recorded in the list one by one. Combining this with the actual frame rate parameters of the video, it locates the corresponding frame index range in the original video sequence. The system default frame rate is 30fps, meaning 30 frames per second. When the start and end times in the time list are between 2.5 and 5.5 seconds, the video segment localization submodule calculates the corresponding frame index as frames 75 to 165 by multiplying the time by the frame rate. This range contains a total of 91 frames, and this frame sequence is the video segment to be extracted. The module accesses the frame data of the original video stream through the buffer, reading consecutive image frames within this frame range to ensure image continuity and consistency with the timeline. Each image frame is identified by an image number, such as F075 to F165. The system then writes this sequence into the fluctuation frame segment image sequence set, recording the index range of the current sequence in the original video and attaching timestamp information. For example, in a construction monitoring video, a paver is found to exhibit significant movement changes within the 3.0s to 5.0s interval, corresponding to frame segments F090 to F150. The segment localization submodule will extract this image frame sequence, generate a fluctuating frame segment image sequence set, and record the frame range of this segment as [F090, F150]. To control the efficiency of video data reading, the system sets a minimum frame segment length threshold of 15 frames. This value is set based on the shortest continuous frame count of the target action in the test scenario. Tests have shown that at least 15 frames (0.5 seconds) of continuous action are required to ensure effective recognition. Therefore, frame segments shorter than 15 frames are ignored and not extracted. If a certain frame segment has a start and end frame count of F045 to F057, which is only 13 frames, it does not meet the minimum length threshold and will not be written into the fluctuating frame segment image sequence set.

[0063] The region number statistics submodule calls the fluctuating frame segment image sequence set, detects the region number corresponding to each frame in the image region, counts the cumulative number of occurrences of the number in the frame segment, extracts the region number with the highest frequency, and determines whether there is a shot switching mark based on the change order of the region number in consecutive frames, and generates the region number stability state.

[0064] The region numbering and statistics submodule extracts each frame segment from the fluctuating frame segment image sequence set. It processes all image frames within a segment sequentially. Each frame has already been preprocessed to define the target region and assigned a number. This module extracts the region numbers appearing in the image frame by frame and counts the cumulative occurrences of each number throughout the entire frame segment. For example, in the F100 to F130 frame segment, region number "R08" appears in 28 frames, number "R09" appears in 20 frames, while number "R10" appears only in 7 frames. Therefore, "R08" is considered the dominant region number for this frame segment due to its highest frequency. The system sets the numbering threshold to 60% of the frame segment length. That is, if a region number appears more than 60% of the total number of frames, it is considered the dominant number. This threshold is set based on statistical analysis of actual video scenes. In 100 sample segments, the average coverage rate of effective target region numbers in the frame segment was found to be 65%. Therefore, using 60% as the dominant number threshold is representative. Subsequently, the module analyzes the sequential changes of the numbers within the frame segments to determine whether a shot switching behavior exists. If the same number disappears after a certain frame, and the number abruptly changes to a previously unseen number in the next frame, it may indicate a shot switching. For example, if the number "R05" exists continuously from F110 to F118, but is suddenly replaced by "R12" in F119, and "R12" continues to exist from F120 to F130, then F119 is recorded as the starting frame of the shot switching, and a shot switching marker is generated. The system sets the threshold for continuous frame switching judgment to 2 frames. That is, if a number is suddenly replaced within two frames, and the preceding and following numbers are different, it is marked as a potential switch. This threshold is verified through experiments. In 200 segments of real scene data, the impact of different frame number changes on the accuracy of shot switching recognition was tested, and 2 frames was found to be the optimal threshold, generating a region number stability state.

[0065] The segment numbering and grouping submodule filters out frame segment numbers with shot switching markers based on the stability status of the region numbering, calls the region numbers counted in the frames that have not undergone shot switching, and performs clustering and merging based on the numbering fit degree and spatial adjacency relationship to obtain the image segment number set.

[0066] The segment numbering and grouping submodule first calls the region numbering stability status to filter all frames, removing those with shot transition markers. Only segments whose region numbers appear consistently throughout the entire frame segment are retained for further processing. Then, spatial adjacency detection is performed on each main region number analyzed by the submodule, grouping spatially close numbers into the same segment group. The merging process considers two parameters: number fit and positional adjacency. The fit represents the percentage of frames where two numbers overlap within a segment. For example, if numbers "R07" and "R08" appear simultaneously in 22 out of 30 frames, the fit is 22 / 30. =73.3%, the system sets the fitting degree merging threshold to 70%. If it is less than this value, it will not be merged. This threshold was set through experiments. In 60 construction monitoring data, it was found that the fitting degree between stable targets in the same scene is generally above 72%. Taking 70% can filter out short-term coexistence areas and avoid excessive merging. The positional adjacency value is based on the pixel spacing of the target boundary in the image. If the distance between the center points of two numbered targets on the image plane is less than 100 pixels, they are considered to be spatially adjacent. 100 pixels is an empirical value. In actual scenarios, the center distance between adjacent targets (such as paving vehicles and road rollers) is usually between 80 and 120 pixels. Therefore, the median value of 100 pixels was selected. For example, the numbers "R02" and "R03" exist simultaneously in 24 frames from frame F060 to F090, with a fitting degree of 24 / 31=77.4%, and the average center distance in the image is 95 pixels, which meets the merging condition. They are merged into the segment number "P002", and finally the image segment number set is obtained, such as {P001, P002, P003}, each number pointing to a relatively independent and stable image scene sequence.

[0067] Please see Figure 4 The frequent fragment aggregation and localization module includes:

[0068] The numbered segment reading submodule calls the segment corresponding to the number in the image segment number set, locates the segment frame in the video data, extracts the lens direction attribute and start and end time labels within the frame segment, establishes a mapping index between number, direction and time, and generates a number attribute mapping table.

[0069] The numbered segment reading submodule calls each numbered item in the image segment number set and sequentially reads the corresponding segment frame mapping relationship. During the reading process, it first locates the corresponding start and end frame numbers based on the number index (e.g., P001), and calculates the actual shooting time stamp based on the system's standard frame rate (default is 30fps). For example, if number P001 corresponds to frame segment F060 to F090, the start frame time is F060÷30=2.0 seconds, and the end frame time is F090÷30=3.0 seconds. The duration corresponding to this number in the video sequence is 1 second. Subsequently, the submodule extracts the corresponding frame segment from the video buffer. The system generates frame image data, and for each frame read, it includes a lens orientation attribute value. The orientation value is an integer between 0° and 359°, in degrees, representing the lens orientation. The orientation value is acquired by the gyroscope sensor of the device's internal IMU unit. The shooting orientation is recorded at the frame-by-frame granularity, and each frame image carries an orientation field. When loading frame segments, the orientation information is read synchronously to form a frame-level orientation sequence. The starting value of this sequence is the starting orientation of the numbered segment, and the ending frame orientation is the ending orientation. If the orientation of frame F060 in number P001 is 275° and the orientation of frame F090 is 278°, then the orientation range of segment P001 is [275°, 278°]. To improve processing efficiency, the system sets the orientation recording accuracy threshold to ±1°. This value comes from the angular velocity stability test of the IMU chip under static conditions. The maximum deviation of the orientation value change in 10 video segments is 0.8°. Setting ±1° can filter out errors caused by measurement jitter. Finally, the module constructs a complete mapping index of number, direction, and time based on the above information. Each number is bound to the start frame, end frame, direction interval, and time interval, generating a number attribute mapping table and storing it in the attribute cache pool for reference by subsequent modules.

[0070] The lens attribute extraction submodule extracts the lens direction value and time tag in the number group according to the number attribute mapping table, arranges the numbered frame segments in time order, extracts the lens direction sequence, and obtains the numbered lens direction sequence.

[0071] After obtaining the number attribute mapping table, the lens attribute extraction submodule first extracts the direction value range and time tag value of each number item, and sorts all numbers in ascending order of start time. After sorting, it constructs the number lens direction sequence in sequence. Each number in the sequence is represented as a triplet of "number ID + direction value + start and end time". The numbers in the sequence are arranged in strict ascending order of time. For example, the sequence items are {P001[275°, 2.0s–3.0s], P002[278°, 3.1s–4.1s], P003[280°, 4.3s–5.3s]}. The lens direction is its starting frame direction, and the time tag is the video time interval in which its frame segment is located. During the sequence construction process, the system determines whether there are time or direction conflicts between adjacent numbers. A time conflict refers to an overlap in the start and end times of two numbers. If the start time of number P004 is 5.1s and the end time of P003 is 5.3s, there is a 0.2-second overlap. The module sets a time tolerance threshold of 0.15 seconds. Based on previous analysis, it was found that the number switching interval of most handheld cameras or low-speed tracking shots is between 0.1 and 0.2 seconds. The median value of 0.15 seconds was selected as the boundary. When the actual overlap time exceeds this value, it is marked as a "time conflict," and only numbers with larger frame lengths are retained. Direction conflict is determined based on whether the direction difference threshold exceeds the set range. The system sets the direction change threshold to 15°, which stems from the observation in on-site shooting tests that if the direction difference between adjacent numbers exceeds 15°, it usually reflects a significant change in the shooting target's perspective. The system marks this change as a "direction change." After all number attributes are filtered according to the above rules, a numbered shot direction sequence is generated and recorded in the numbered shot direction index set for use in the next stage of continuous analysis.

[0072] The continuity matching and recognition submodule detects whether the numbering sequence is continuous based on the numbered lens direction sequence, and counts the number of times the direction values ​​are consistent, using the formula:

[0073] ;

[0074] The system calculates the continuity and consistency value of the serial numbers, determines whether it exceeds the set recognition standard threshold, filters out the serial number sequences that meet the conditions, and generates a sequence of continuous shot segments.

[0075] in, This indicates the consistency value of the serial numbers. This indicates the consistency determination result between the current number and its predecessor in the lens direction dimension. and These are the start and end indexes of the current shot segment, respectively. The number of segments that maintain a consistent direction. For the first The starting time frame index of each segment with consistent direction. This is the average of the frame indices of the starting time of this type of segment. This is the starting value of the time segment index variable. and These are the maximum and minimum values ​​of the start time frame indices for all consistent segments. This represents the theoretical upper limit of the normalized directional time concentration coefficient, used to balance the contribution weight of time distribution in the overall calculation;

[0076] Detailed explanation of parameter symbols and structure:

[0077] In this formula:

[0078] : Represents the "numbering continuity consistency value", which is the judgment index output of this submodule;

[0079] : indicates the first The result of the consistency of the direction of the lens number with the previous number is recorded as 1 if the direction difference is less than 7°, otherwise it is recorded as 0, and the value is Boolean.

[0080] and : Indicates the start and end indexes of the current shot segment (e.g., numbers P101 to P106, corresponding to...). );

[0081] : Represents the theoretical maximum value of the normalized directional time concentration coefficient, set to 1.0;

[0082] : Indicates the first segment in the segment with the same direction The start time index (in seconds) of the frame corresponding to each number.

[0083] :express The average value;

[0084] , : The maximum and minimum start time values ​​in segments with consistent direction;

[0085] square root inner term This refers to the standard deviation of time dispersion, which reflects temporal clustering.

[0086] Divide by Normalize the standard deviation to ensure that the entire formula maintains a dimensionless structure;

[0087] External multiplication terms: Combining directional consistency with temporal concentration to form the final continuous consistency result. .

[0088] Variable sampling and parameter assignment instructions (actual data collection):

[0089] Suppose the current analysis segment contains numbered segments:

[0090] P101 (270°), P102 (275°), P103 (276°), P104 (289°), P105 (291°), P106 (292°);

[0091] Assume these numbers correspond to the following frame numbers:

[0092] F303, F304, F305, F306, F307, F308, frame rate 30fps.

[0093] The criterion for judging directional consistency is set at ±7°. By comparing the differences between adjacent directions, we obtain:

[0094] (270→275, difference 5°);

[0095] (275→276, difference 1°);

[0096] (276→289, difference 13°);

[0097] (289→291, difference 2°);

[0098] (291→292, difference 1°);

[0099] then: The first value: ;

[0100] The frame number in the consistency segment is taken as:

[0101] F303, F304, F305, and F308 are converted to timestamps as follows:

[0102] ;

[0103] Calculate the mean:

[0104] ;

[0105] Calculate the square of the deviation:

[0106] ;

[0107] ;

[0108] ;

[0109] ;

[0110] Summation:

[0111] ;

[0112] Square root:

[0113] ;

[0114] Time span:

[0115] ;

[0116] Normalization term:

[0117] ;

[0118] set up Therefore, the second item is:

[0119] ;

[0120] Final calculation:

[0121] ;

[0122] Table 1 Continuity Consistency Parameter Data Table

[0123] ;

[0124] As shown in Table 1, there are a total of 4 numbers with consistent directions. The final calculation of the numbering continuity consistency value is as follows.

[0125] Set the threshold for consecutive number identification as follows: The reference for setting this threshold is as follows: In 3000 sets of construction video test samples, the R value of continuous shooting scenes is mostly distributed between [0.25, 1.0], while the R value of non-continuous or strong jump shots is mostly below 0.2. This threshold is set based on comprehensive evaluation.

[0126] Based on the results of this calculation:

[0127] ;

[0128] Therefore, the current numbering sequence does not meet the condition of continuous shot segments, and the identification result is "non-continuous numbering segment".

[0129] The advantage of the formula is that it introduces consistency in numbering direction. With the dispersion of the start time The standard deviation normalization structure enables the system to make comprehensive judgments on both spatial perspective and temporal clustering dimensions, effectively avoiding misidentification problems caused by short-term directional jitter or dense numbering but shot jumps, and improving the stability and robustness of continuous shot segment extraction.

[0130] Please see Figure 5 The channel segment frequency hopping screening module includes:

[0131] The shot segment extraction submodule extracts the video segments corresponding to the numbers listed in the sequence of consecutive shot segments, parses the frequency point sequence information recorded in the 5G signal tracking nodes on the side of the road, and arranges them into a sequence set according to the number order to obtain the video frequency point sequence set;

[0132] The shot segment extraction submodule reads the corresponding video segments one by one according to the numbered information in the sequence of consecutive shot segments. Each number corresponds to a start frame number and an end frame number. After obtaining the start and end frame indices by calling the number attribute mapping table, the system reads the corresponding frame image content from the original video data. Then, it parses the roadside 5G signal tracking node recording data that overlaps with the time of the frame segment with that number. This data comes from independent frequency point monitoring units installed on both sides of the road. Each unit records the current channel frequency point value every 10 milliseconds and stores it as a time-stamped sequence record file. The system locates the start and end indices of the segment in the frequency point record data according to the time range of the frame segment. For example, the start and end time of number P108 is 4.5 seconds to 5.5 seconds. The system extracts a total of 100 frequency point records within this time period, sorts them by timestamp, and forms a frequency point sequence set. Each item in this sequence contains a time point and a corresponding frequency point value, such as {4.500s, 3512MHz}, {4.510s, 3513MHz}, etc., thus forming a frequency point sequence array corresponding to the number. To ensure synchronization between frequency point recordings and video frames, the system sets a frame-frequency matching tolerance threshold of ±0.05 seconds. That is, if the difference between a frequency point timestamp and the frame time is less than this threshold, it is considered synchronized. This value was obtained through an error alignment experiment between frequency point recordings and video frame encoded timestamps. In the experiment, 1000 synchronized frame points were counted, and it was found that 95% of the matching errors were less than 0.042 seconds. Therefore, 0.05 seconds was selected as the tolerance limit value. Finally, all numbers correspond to a frequency point sequence, forming a complete video frequency point sequence set, which provides input data for the frequency hopping fluctuation analysis module.

[0133] The frequency difference calculation submodule calls the video frequency point sequence set. For the frequency point values ​​recorded at the start and end points of each video segment, it obtains the start and end frequency point values ​​of the video segment, calculates the difference range sequentially, and records the difference data interval using the formula:

[0134] ;

[0135] The frequency hopping difference metric of the video segment is obtained by calculation, and the set of frequency hopping variation amplitude is obtained by combining the difference change amplitude with the frequency point dispersion.

[0136] Where, ΔF k Representative video segment The frequency hopping difference metric. Representative video segment The termination frequency value, Representative video segment The starting frequency value, For video segments The Middle Frame frequency point value, For video segments Total number of frames, Indicates video segment The average frequency value of all frames in the data; where ΔF k Representative video segment The frequency hopping difference metric. Representative video segment The termination frequency value, Representative video segment The starting frequency value, For video segments The Middle Frame frequency point value, For video segments Total number of frames, Indicates video segment The average frequency value of all frames in the middle;

[0137] For the frequency hopping feature extraction process in each video segment, during the frequency difference calculation, the current video segment number is first used as the basis for the process. By determining the corresponding start and end frames of the video, the starting frequency value of the video segment can be obtained. With the termination frequency value Then, the frequency values ​​of all frames within the video segment are read. Total number of frames And based on it, calculate the frequency average. This value is obtained by calculating the arithmetic mean of all frequency values ​​within the segment, i.e., by performing the operation. Therefore, the first term in the above formula is adopted. The first term represents the overall amplitude of frequency hopping, used to reflect the absolute deviation of the start and end frequency values, while the second term... This indicates the dispersion of frequency points within the current video segment; a larger value indicates more severe frequency fluctuations. The sum of the two terms is then used to calculate the square root of the result, which forms the final frequency hopping difference metric. This item has a Hz dimension, which can be used as a frequency point stability evaluation index. When extracting values, the parameters are assigned based on the actual sampled frames. The following provides a detailed parameter description and example calculation.

[0138] In this example, the number selected is The video segment was used as a sample, and its starting frequency value was... The termination frequency value is The total number of frames in the video segment is The frequency point values ​​corresponding to each frame are read sequentially as follows: , , , , According to the formula for the average frequency point, we have:

[0139] ;

[0140] Then, the following operation is performed on the squared difference term:

[0141] The first term is the squared term of the frequency jump variation:

[0142] ;

[0143] The second term, the sum of squares of frequency dispersion, is:

[0144] ;

[0145] ;

[0146] Finally, substituting into the formula, we get:

[0147] ;

[0148] The result indicates that the frequency jump metric for the current video segment number 3 is 25.89MHz. If the system stability threshold is set to 30MHz, then this video segment is determined to be within the acceptable range of frequency change. The participating values ​​and intermediate calculation results are shown in Table 2.

[0149] Table 2: Video Segment Frequency Point Parameters and Calculation Results

[0150] ;

[0151] As shown in Table 2, the frequency hopping difference of this video segment is below the set threshold, which meets the stable screening condition.

[0152] The advantage of the formula is that by simultaneously calculating the difference between the start and end frequency points and the dispersion of frequency fluctuations within the segment, it comprehensively measures the frequency hopping amplitude and stability, thereby improving the sensitivity and accuracy of frequency hopping change detection. This enables the system to capture both large sudden changes and identify small high-frequency disturbances, thus enhancing the robustness and resolution of video segment screening as a whole.

[0153] The frequency hopping stability filtering submodule compares the frequency hopping difference metric of video segments with the set signal frequency hopping stability threshold one by one based on the frequency hopping change amplitude set, and filters out all video segment numbers whose frequency hopping difference metric is less than the signal frequency hopping stability threshold to obtain a set of signal stable video segment numbers;

[0154] The frequency hopping stability screening submodule reads the frequency change information of each video segment numbered by segment according to the frequency hopping change amplitude set, and compares its frequency hopping difference metric with the signal frequency hopping stability threshold set by the system. The stability threshold is set to ±5MHz. The selection basis is that the average frequency fluctuation in 1800 video segments recorded by the roadside 5G communication equipment under real working conditions is 3.9MHz. In 90% of stable scenarios, the frequency hopping difference is less than 5MHz. More than 10% of high-frequency fluctuations occur in scenarios with shortened line of sight, severe signal interference, or antenna obstruction. Therefore, the frequency hopping difference judgment standard is set to 5MHz. When the frequency hopping difference metric of a certain number is less than this value, the segment is considered to be frequency stable. The system introduces a dual judgment rule: if the frequency difference is less than 5MHz, the fluctuation amplitude must not exceed 15MHz. Otherwise, it is considered that although the start and end frequencies are similar, the fluctuations in the middle are severe and do not constitute a truly stable signal segment. For example, the frequency hopping difference of segment P110 is 3MHz, but its fluctuation amplitude is 17MHz, exceeding the system's set amplitude limit of 15MHz, so it is not included in the stable segment. Conversely, segment P111 has a difference of 4MHz and a fluctuation amplitude of 12MHz, meeting both conditions, and the system determines it to be a stable signal segment. All the final selected stable segment numbers are uniformly written into the signal stable video segment number set for subsequent use in constructing the signal-action behavior association feature table.

[0155] Please see Figure 6 The video segment scheduling trigger module includes:

[0156] The timestamp extraction submodule calls the segment numbers in the signal-stable video segment number set, reads the first frame timestamp and last frame timestamp in the video segment corresponding to each number, and performs number mapping processing on the time information to generate a video segment time index table.

[0157] The timestamp extraction submodule processes the numbered items in the stable video segment number set sequentially, reading the original video segment frame index data corresponding to each numbered item. It then locates the actual frame image based on the first and last frame numbers of that segment. Simultaneously, it extracts the system timestamp field attached to the frame header during frame image reading. This field is a floating-point time value in "seconds.milliseconds" format, uniformly represented using the UTC time zone standard. To ensure the accuracy of the extracted time, the system sets the timestamp extraction error tolerance threshold to ±0.01 seconds. This value comes from the statistical results of encoding time errors measured in 1200 video samples in the test environment, where 99% of the first frame time errors are within 0.009 seconds. Therefore, 0.01 seconds is selected as the precision control range. When extracting the timestamp, if an error is found between frames... If there is a discrepancy in the timing logic, the system will automatically perform padding and shifting operations. For example, if the frame segment corresponding to number P301 is F600 to F630, the corresponding duration in a 30fps video should be 1.0 second. The system timestamp of the first frame is read as 20.010s, and the system timestamp of the last frame is 21.009s. The system checks that the difference between the two values ​​is 0.999 seconds, which is within the error threshold tolerance range, so it is directly recorded as [20.010s–21.009s]. Otherwise, the offset correction program will be triggered to reread. All the first and last timestamp data extracted by number are sorted and organized according to number and bound to the number index to generate a video segment time index table. The structure fields of this table include number, start frame, end frame, start time, end time, duration, etc., which are used for subsequent time matching operations with channel information.

[0158] The task information binding submodule extracts the video task identifier field and video channel binding information for each time interval based on the video segment time index table, reads the task sequence value in chronological order, and establishes a binding record index to obtain the task channel binding structure.

[0159] The task information binding submodule calls the time interval data in the video segment time index table and integrates it with the existing task information table and channel record table in the road video dispatch management system. Each task record in the task information table includes fields for task number, task start time, end time, channel number, and task type. The module reads the numbered time intervals in the time index table one by one and performs a complete containment match according to the time interval to determine whether the current numbered time interval is completely wrapped by a certain task time interval. The matching judgment threshold is set to a coverage ratio of 80%, that is, when the overlap length between the numbered time interval and the task time interval divided by the total length of the numbered time interval is greater than or equal to 0.8, the system determines that the number belongs to the task segment. This coverage ratio is determined by analyzing the overlap between the actual execution waveform of the task and the video segment data in 600 road operation dispatch scenarios. According to the statistics from the coverage experiment, the numbering with a coverage rate exceeding 0.8 maintains temporal consistency with the task number, which conforms to the operational logic. Therefore, it is set as a reasonable minimum effective coverage judgment standard. For example, the time period of number P302 is [22.000s–23.000s], and the time period of a certain task T063 is [21.900s–23.200s]. The overlap interval is 1.000s, the total duration of the number is 1.000s, and the overlap ratio is 100%, which meets the conditions. The system binds task T063 with number P302, and at the same time extracts the task binding channel field CH-09 and writes it into the binding record. All task numbers, channel numbers and lens numbers that pass the binding rules form a one-to-one relationship and are arranged in chronological order to form a task channel binding structure, which is used for the number mapping construction of subsequent scheduling sequences.

[0160] The scheduling mapping generation submodule matches the start and end boundaries of adjacent task number combinations according to the task sequence value, first and last frame timestamps and channel information in the task channel binding structure, sets the scheduling start and end numbers, establishes a bidirectional mapping relationship between the shot number and the task scheduling sequence, generates scheduling task shot mapping items, and obtains the road construction video inspection and data analysis results.

[0161] The scheduling mapping generation submodule receives the task channel binding structure and sorts all numbered entries according to task number and time order. The system sets the numbering order to ascending order of start time. When the start times of the numbers are exactly the same, they are prioritized to be sorted in ascending order of channel number to avoid concurrent scheduling conflicts. After sorting, the module processes adjacent numbered records in pairs, comparing the time boundary between the current task number and the next task number one by one, and determining whether they constitute a continuous scheduling relationship. The judgment rule sets the maximum time gap threshold to 0.2 seconds. That is, if the interval between the end time of the previous number and the start time of the next number is less than or equal to 0.2 seconds, it is considered a schedulable continuous boundary pair. This value comes from the statistical data of crew task switching time in the job scheduling mechanism. In 3200 actual task switching records, the time interval does not exceed 0.19 seconds in 90% of the scenarios, hence the setting. The boundary value is 0.2 seconds to ensure that the system can identify continuous scheduling relationships without misjudging short delays. For example, if the time period of number P303 is [24.000s–25.000s] and the time period of P304 is [25.050s–26.100s], with an interval of 0.050s, which is less than 0.2 seconds, the two numbers can be used to establish a scheduling mapping. The system uses P303 as the starting number and P304 as the ending number to record the task number pair T064→T065 and the channel number pair CH-08→CH-08. The system also writes this mapping into the scheduling task shot mapping item. Finally, all mapping groups that satisfy the conditions of continuous time and logically increasing numbering form a scheduling task shot mapping relationship structure, which supports bidirectional index retrieval between shot segments and task scheduling. This is used for construction task execution trajectory reconstruction and channel scheduling efficiency analysis, as well as road construction video inspection and data analysis results.

[0162] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A 5G-based road construction video inspection and data analysis system, characterized in that, The system includes: The video frame sequence extraction module acquires 5G video of the road paving section and the intersection construction section, extracts the spatial contour of the target in the continuous frames, calculates the edge displacement and overlap ratio, determines whether the offset threshold is exceeded, and outputs a list of motion fluctuation frame times. The operation behavior fluctuation recognition module locks the video segment according to the action fluctuation frame time list, counts the image region numbers that appear most frequently, and if there is no shot switching mark, it clusters and groups them to obtain a set of image segment numbers; The frequent segment aggregation and positioning module calls the image segment number set, reads the shot direction and time tag, calculates the proportion of number continuity and direction consistency, and outputs the sequence of continuous shot segments. The channel segment frequency hopping screening module reads the sequence of the continuous shot segments, obtains the 5G frequency point changes of the corresponding video segments, calculates the frequency point difference and compares it with the frequency hopping stability value, and outputs a set of signal-stable video segment numbers. The video segment scheduling trigger module calls the set of signal-stable video segment numbers, extracts the timestamp and task field, combines the task sequence value to establish a scheduling index, records it as a camera mapping item, and outputs the road construction video inspection and data analysis results.

2. The 5G-based road construction video inspection and data analysis system according to claim 1, characterized in that, The motion fluctuation frame time list includes frame segment start and end time indexes, target edge displacement values, and image overlap ratio values. The image segment number set includes occurrence region numbers, region cluster numbers, and shot switching identifiers. The continuous shot segment sequence list includes shot number sequence, shot direction consistency, and number continuity index. The signal stable video segment number set includes frequency point variation range, frequency hopping stability threshold, and stable segment number. The road construction video inspection and data analysis results include shot start and end timestamps, task identifier codes, video channel information, and scheduling order indexes.

3. The 5G-based road construction video inspection and data analysis system according to claim 2, characterized in that, The video frame sequence extraction module includes: The video data receiving submodule acquires the image frame sequence collected by 5G high-definition video monitoring points deployed in the road paving section and the intersection construction section, detects the position of the target object in the continuous frame, extracts the spatial contour boundary of the target in each frame, and generates a set of target frame contour regions. The contour edge calculation submodule, based on the target frame contour region set, compares the target edge coordinates in adjacent frames, calculates the average displacement distance value and the contour overlap ratio value, compares the displacement distance value with the cumulative frame edge offset threshold, obtains the frame index number that is greater than the cumulative frame edge offset threshold, and obtains the image frame displacement offset index list. The frame sequence number generation submodule calls the image frame displacement offset index list, filters the frame index numbers of continuously changing segments, matches them to a set time window range, establishes a time axis mapping relationship, and generates a motion fluctuation frame time list.

4. The 5G-based road construction video inspection and data analysis system according to claim 3, characterized in that, The operational behavior fluctuation identification module includes: The video segment localization submodule matches the corresponding frame image interval in the original video sequence according to the frame start and end time index in the motion fluctuation frame time list, extracts the continuous image frame sequence within the frame segment interval, records the frame index range of the sequence in the video, and generates a set of fluctuation frame segment image sequences. The region number statistics submodule calls the fluctuating frame segment image sequence set, detects the region number corresponding to each frame in the image region, counts the cumulative number of occurrences of the number in the frame segment, extracts the region number with the highest frequency, and determines whether there is a shot switching mark based on the change order of the region number in consecutive frames, and generates the region number stability state. The segment numbering and grouping submodule filters out frame segment numbers with shot switching marks based on the stability status of the region numbers, calls up the region numbers counted in the frames that have not undergone shot switching, and performs clustering and merging based on the number fitting degree and spatial adjacency relationship to obtain the image segment number set.

5. The 5G-based road construction video inspection and data analysis system according to claim 4, characterized in that, The frequent fragment aggregation and localization module includes: The numbered segment reading submodule calls the segment corresponding to the number in the image segment number set, locates the segment frame in the video data, extracts the lens direction attribute and start and end time labels within the frame segment, establishes a mapping index between number, direction and time, and generates a number attribute mapping table. The lens attribute extraction submodule extracts the lens direction value and time tag in the number group according to the number attribute mapping table, arranges the numbered frame segments in time order, extracts the lens direction sequence, and obtains the numbered lens direction sequence. The continuity matching and recognition submodule detects whether the numbering order is continuous based on the numbered shot direction sequence, counts the numbering times the direction values ​​are consistent, calculates the continuity consistency value of the numbers, determines whether it is higher than the set recognition standard threshold, filters the numbering sequences that meet the conditions, and generates a sequence of continuous shot segments.

6. The 5G-based road construction video inspection and data analysis system according to claim 5, characterized in that, The channel segment frequency hopping screening module includes: The shot segment extraction submodule extracts the video segments corresponding to the numbers listed in the sequence of the continuous shot segments, parses the frequency point sequence information recorded in the 5G signal tracking node on the side of the road, and arranges them into a sequence set according to the number order to obtain the video frequency point sequence set; The frequency difference calculation submodule calls the video frequency sequence set, obtains the start and end frequency values ​​of the video segment for the frequency hopping frequency values ​​recorded at the start and end points of each video segment, calculates the difference range and records the difference data interval in sequence, calculates the frequency hopping difference metric value of the video segment, and obtains the frequency hopping change amplitude set by combining the difference change amplitude and the frequency dispersion. The frequency hopping stability filtering submodule compares the frequency hopping difference metric of the video segment with the set signal frequency hopping stability threshold one by one based on the frequency hopping variation amplitude set, and filters out all video segment numbers whose frequency hopping difference metric is less than the signal frequency hopping stability threshold to obtain a set of signal stable video segment numbers.

7. The 5G-based road construction video inspection and data analysis system according to claim 6, characterized in that, The video segment scheduling trigger module includes: The timestamp extraction submodule calls the segment numbers in the signal-stable video segment number set, reads the first frame timestamp and the last frame timestamp in the video segment corresponding to each number, and performs number mapping processing on the time information to generate a video segment time index table. The task information binding submodule extracts the video task identifier field and video channel binding information for each time interval according to the video segment time index table, reads the task sequence value in chronological order, and establishes a binding record index to obtain the task channel binding structure. The scheduling mapping generation submodule matches the start and end boundaries of adjacent task number combinations according to the task sequence value, first and last frame timestamps and channel information in the task channel binding structure, sets the scheduling start and end numbers, establishes a bidirectional mapping relationship between the shot number and the task scheduling sequence, generates scheduling task shot mapping items, and obtains the road construction video inspection and data analysis results.

8. The 5G-based road construction video inspection and data analysis system according to claim 5, characterized in that, The process of obtaining the continuity consistency value of the number is to determine whether the direction values ​​of two adjacent shots in the numbered shot direction sequence are consistent, and to use a set weighting ratio to statistically accumulate the number of consistency. In the process of determining whether the value is higher than the set recognition standard threshold, the judgment is made based on the ratio between the number of times the direction value is consistent and the total number of the numbered lens direction sequence, combined with the preset continuity evaluation rules.

9. The 5G-based road construction video inspection and data analysis system according to claim 6, characterized in that, In the process of obtaining the video frequency point sequence set, each video segment corresponding to each number is limited to include a start frequency point value and an end frequency point value, and both of them have time tags; In the process of comparing the frequency hopping difference metric of the video segment with the set signal frequency hopping stability threshold one by one, the signal frequency hopping stability threshold is determined based on the dispersion of the frequency hopping change amplitude set, and the set of signal stable video segment numbers is verified to be continuously arranged on the time label.

Images