Construction site safety monitoring system based on drone vision

Abstract

The present invention discloses a construction site safety monitoring system based on drone vision. The present invention relates to the technical field of construction site safety monitoring and solves the problem of being unable to uniformly identify whether relevant construction workers are wearing safety helmets and whether their safety helmets are worn in compliance with standards. The present invention uses a drone to monitor the construction area of ​​the construction site in real time, and determines the monitoring screen in real time, and then selects a face image from the monitoring screen based on relevant extracted features, and then performs a preliminary analysis on the face image. By locking the center point and the point to be processed, it is identified whether the mean angle of the connecting line within the contour is consistent, and a preliminary judgment can be made as to whether the corresponding construction worker's helmet is worn abnormally, and a relevant judgment signal is generated for timely display. The monitoring method is more comprehensive and can achieve better construction site safety supervision effects.

Description

Technical Field

[0001] The present invention relates to the technical field of construction site safety monitoring, and in particular to a construction site safety monitoring system based on drone vision. Background Art

[0002] With the rapid development of science and technology, the identification and prevention of safety hazards have become an indispensable component in all fields. In particular, in highway construction, the timely detection and resolution of safety hazards are crucial for ensuring personnel safety and minimizing economic losses. Against this backdrop, drone airports, as a fast and flexible inspection tool, demonstrate unique advantages and potential in the highway sector.

[0003] Through automated control and intelligent management, the drone airport can achieve rapid scheduling and periodic inspections, perfectly adapting to the requirements for safety inspections during highway construction.

[0004] Equipped with high-definition cameras and advanced sensors, drones are able to capture details that are difficult to observe on the ground.

[0005] The application with publication number CN113296427B discloses a construction site safety monitoring system based on the Internet of Things, which relates to the field of construction safety monitoring technology; it includes an environmental analysis module, a database, an environmental safety assessment module and a behavior analysis module; the environmental analysis module is used to analyze environmental parameter data to obtain a first safety state; the environmental safety assessment module is used to obtain the current environmental video information of the construction workers in real time; and evaluate the environmental safety in combination with the working environment safety status data in the database to obtain a second safety state; the behavior analysis module is used to identify the actions or movement behaviors of the construction workers to obtain the identification results of illegal actions or movements; and perform safety analysis based on the identification results to obtain a third safety state; the present invention combines the first safety state, the second safety state and the third safety state to comprehensively analyze the environment in which the construction workers are located, and issues reminders based on the analysis results, saving human resource consumption and improving safety.

[0006] Most of the safety hazards at construction sites are caused by the relevant construction workers not wearing safety helmets correctly. However, the existing safety monitoring system can only determine whether the relevant construction workers are wearing safety helmets and issue safety warnings through preset image feature recognition methods. However, this method is relatively one-sided and cannot uniformly identify whether the relevant construction workers are wearing safety helmets and whether their safety helmets are worn in compliance with standards, resulting in poor comprehensiveness of its safety monitoring. Summary of the Invention

[0007] In response to the shortcomings of the existing technology, the present invention provides a construction site safety monitoring system based on drone vision, which solves the problem of not being able to uniformly identify whether relevant construction workers are wearing safety helmets and whether their safety helmets meet the standards.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: a construction site safety monitoring system based on drone vision, comprising:

[0009] The drone's visual system controls the drone's flight based on a preset trajectory and captures the construction scene in real time. The drone's internal visual system has its internal shooting angles pre-set before flight.

[0010] The image primary processing end locks the relevant people appearing in the construction picture based on the set human features, and then extracts the associated images of the facial areas of the relevant people based on the preset face template. After adjusting the clarity of the associated images, the images to be verified are obtained and transmitted to the image primary analysis end;

[0011] The initial image analysis end identifies the overall relevant contours of the image to be verified based on the processed image to be verified, and by eliminating the facial area in the image to be verified, determines whether the person associated with the image to be verified has worn any abnormal clothing. The specific method is as follows:

[0012] Based on the facial region marked in the image to be verified, the center point of the facial region is locked, and a set of vertical lines passing through the center point and perpendicular to the horizontal base plane is constructed. The horizontal base plane is a preset horizontal plane. The intersection of the vertical lines and the overall outer contour of the image to be verified is locked, and the intersection point is marked as the processing point;

[0013] The same plane where the center point and the point to be processed are located is marked as the selected surface, the part of the contour inside the selected surface is marked as the characteristic contour, and the characteristic contour is divided into several characteristic points at equal intervals;

[0014] The points to be processed divide this feature contour into a left contour and a right contour, and prioritize identifying the associated angle between adjacent points in the left contour and the horizontal plane. Connect the adjacent feature points and translate the line until it intersects the horizontal plane. Lock the angle and mark it as J. i , where i represents the angle between the lines of different adjacent points, and the angles J determined by the left contour are i Perform mean processing to determine the left contour mean JZ1;

[0015] Then process the right contour in the same way to determine the right contour mean JZ2;

[0016] If JZ1≠JZ2, it means that the helmet is worn abnormally or the helmet is not worn, and subsequent relevant analysis and processing are required;

[0017] If JZ1=JZ2, it means the helmet is worn normally and no further processing is required;

[0018] The secondary adjustment recognition end is used to identify the wearer's abnormalities in the image to be verified. The process is to adjust the points to be processed, which are marked by the characteristic contours in the image to be verified. The process then identifies whether the points still have abnormalities after the adjustment. If they still have abnormalities, the point angle processing end is executed. The specific method is as follows:

[0019] Based on the points to be processed marked in the characteristic contour, the points to be processed are moved on the characteristic contour, and it is recorded whether the left contour mean JZ1 and the right contour mean JZ2 satisfy the following conditions in each different movement process: JZ1 = JZ2:

[0020] If so, a non-compliant wearing signal is generated through the signal generating terminal and the relevant information is displayed directly;

[0021] If it does not exist, it means that there is still an abnormality in the associated personnel, and the point angle processing end is executed for correlation analysis;

[0022] The point angle processing end, based on the characteristic contour determined in the image to be verified, identifies several groups of point lines determined within this characteristic contour. Starting from the edge points of the characteristic contour, it sequentially determines the angle parameters between adjacent lines and determines the angle sequence. Then, by analyzing the proportion of the same angle in the angle sequence, it assesses whether the relevant personnel are wearing safety helmets. The specific method is as follows:

[0023] Based on a number of feature points divided within the feature contour, a group of feature points at the edge of the feature contour are selected as initial points, a line connecting the initial point and its adjacent feature points is determined as a front line, a group of lines subsequent to the front line is then used as a back line, the intersection of the front line and the back line is used as an extension point, the front line is extended, and the angle between the extension line and the back line is locked as a first set of angles;

[0024] Then, the back line is used as the front line, and the back line of the previous line is determined, and the second set of angles is locked. Similarly, the angles between subsequent adjacent lines are confirmed one by one, and the confirmed angles are sorted according to the confirmation order to confirm the angle sequence;

[0025] Identify the angles with the same value that appear in the angle sequence. When the number of angles with the same value is ≥5, these angles with the same value are calibrated as standard angles.

[0026] Identify the total proportion of standard angles in the angle sequence, Zb, where Zb = the total number of standard angles / the total number of angles in the angle sequence. If Zb ≥ Y1, a non-compliant wearing signal is generated through the signal generation end and directly displayed. Y1 is a preset value.

[0027] If Zb<Y1, a no-wearing signal is generated through the signal generating terminal and displayed directly;

[0028] When the number of angles with the same value is less than 5, no calibration of the standard angle is performed.

[0029] The present invention provides a construction site safety monitoring system based on drone vision. Compared with the existing technology, it has the following advantages:

[0030] The present invention uses a drone to monitor the construction area of ​​a construction site in real time, determines the monitoring screen in real time, selects a face image from the monitoring screen based on relevant extracted features, and then performs a preliminary analysis of the face image. By locking the center point and the point to be processed, it is identified whether the mean angles of the connecting lines within the contour are consistent. It can preliminarily determine whether the corresponding construction worker's helmet is worn abnormally, and generate relevant judgment signals for timely display. The monitoring method is more comprehensive and can achieve better construction site safety supervision effects.

[0031] In order to more accurately identify whether the relevant personnel are wearing safety helmets correctly, by identifying whether the means of the left and right contours are equal after adjustment, it is assessed whether they need to proceed to the next step of analysis, and then the angle sequence is determined by analyzing the relevant angles of the corresponding contours. By analyzing the proportion of the same angle in the angle sequence, it is assessed whether the relevant personnel are wearing safety helmets. The evaluation method of sequential analysis and processing can achieve better safety monitoring effects, and can fully identify whether the relevant personnel are wearing safety helmets and whether the safety helmets are worn in compliance with the standards, which can achieve better safety monitoring effects and fully guarantee the safety of construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a schematic diagram of the principle framework of the present invention;

[0033] Figure 2 Schematic diagram of the division of the external contour of the present invention. DETAILED DESCRIPTION

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0035] Example 1

[0036] See also Figure 1 The present application provides a construction site safety monitoring system based on drone vision, including a drone vision end, an image primary processing end, an image primary analysis end, a secondary adjustment and recognition end, a point angle processing end, and a signal generation end, wherein the drone vision end, the image primary processing end, the image primary analysis end, the secondary adjustment and recognition end, the point angle processing end, and the signal generation end are electrically connected in sequence from an output node to an input node;

[0037] Among them, the drone vision end controls the drone to fly according to the preset trajectory based on the drone's operation, and takes real-time photos of the construction scene in the construction area. The internal visual end of the drone has pre-set the internal shooting angle before the flight, and transmits the real-time construction scene to the image primary processing end. Specifically, each different construction area has been pre-marked in the drone's set trajectory. When the relevant drone reaches the corresponding mark, it will start to take pictures of the construction area;

[0038] Among them, the image initial processing end locks the relevant people appearing in the construction picture based on the set human features, and then extracts the associated images of the facial parts of the relevant people based on the preset face template, and obtains the image to be verified after the clarity adjustment processing of the associated images, and transmits it to the image initial analysis end: Specifically, since the technical content of locking the people and the faces of the people in the image belongs to the existing technology, it will not be described in detail here. When selecting the people, target detection is generally performed by combining pre-set features with a classifier such as a support vector machine (SVM); by sliding a window on the image, extracting features in the window and classifying them to determine whether the target person is included, and then using facial recognition technology to extract facial features of the people in the picture, such as the shape and position of the eyes, nose, mouth, etc., or by using existing facial recognition algorithms and software, such as the face detection and recognition module in OpenCV, etc., the relevant people in the picture and the relevant faces of the corresponding people can be identified and obtained. The obtained associated images are then subjected to image sharpening and denoising processing (that is, clarity adjustment processing) to obtain the final image to be verified;

[0039] The initial image analysis end identifies the overall relevant contours of the image to be verified based on the processed image to be verified, and by eliminating the facial area in the image to be verified, determines whether the person associated with the image to be verified is wearing abnormal clothing. The specific method of determining whether the associated person is wearing abnormal clothing is as follows:

[0040] Based on the facial area marked in the image to be verified, the center point of the facial area is locked (the center point can be locked by determining the center point coordinates based on the coordinate point method. For example, a surface is divided into several points, and each point has a related coordinate point in the two-dimensional coordinate system. The corresponding center point can be determined by averaging the several coordinate points. The center point can determine the corresponding center point coordinates. Based on these coordinates, the center point of the corresponding facial area can be locked). A set of vertical lines passing through the center point and perpendicular to the horizontal base plane is constructed. The horizontal base plane is a preset horizontal plane (which can be understood as a related plane parallel to the ground). The intersection of the vertical line and the overall external contour of the image to be verified is locked, and the intersection is marked as the point to be processed;

[0041] The same plane where the center point and the point to be processed are located is calibrated as the selected surface, and the part of the contour inside the selected surface is calibrated as the feature contour. The feature contour is divided into several feature points at equal intervals (because the line is composed of points, the corresponding feature contour can be divided into several points);

[0042] The points to be processed divide this feature contour into a left contour and a right contour, and prioritize identifying the associated angle between adjacent points in the left contour and the horizontal plane. Connect the adjacent feature points and translate the line until it intersects the horizontal plane. Lock the angle and mark it as J. i , where i represents the angle between the lines of different adjacent points, and the angles J determined by the left contour are i Perform mean processing to determine the left contour mean JZ1;

[0043] Then process the right contour in the same way to determine the right contour mean JZ2;

[0044] If JZ1=JZ2, it means the helmet is worn normally and no further processing is required. If JZ1≠JZ2, it means the helmet is worn abnormally or the helmet is not worn. Subsequent relevant analysis and processing are required to determine the specific cause of the abnormality.

[0045] Specifically, when the associated person is not wearing a helmet, the outline generated on his or her outside is the associated outline generated by the corresponding person's hair. Then such outline is relatively messy, and the angle data generated is definitely inconsistent. If a helmet is worn but not worn standardly, the corresponding mean of the angle data generated is also inconsistent. In order to better conduct identification analysis and determine the specific identification results, subsequent association analysis is required to determine whether such related personnel have safety hazards and conduct comprehensive evaluation.

[0046] Among them, the secondary adjustment recognition end, for the case where the person associated with the image to be verified has abnormal wearing, adjusts the processing point marked by the characteristic contour in the image to be verified, and identifies whether it still has abnormalities after the relevant adjustment. If it is still abnormal, the point angle processing end is executed. The specific method of identification is:

[0047] Based on the points to be processed calibrated within the characteristic contour, the points to be processed are moved on the characteristic contour, and it is recorded whether the left contour mean JZ1 and the right contour mean JZ2 satisfy the following condition: JZ1 = JZ2 during each different movement process. If so, a wearing failure signal is generated through the signal generation end and directly displayed.

[0048] If not, it means that there is still an abnormality in the associated personnel, and the point angle processing end is executed to perform correlation analysis to assess whether the corresponding associated personnel are wearing the helmet in the wrong position or not wearing the relevant safety helmet. Based on the specific assessment and analysis results, a comprehensive assessment is carried out in time, and relevant processing signals are generated in time for real-time display to ensure the construction safety of relevant construction personnel in the construction area;

[0049] Specifically, if the helmets worn by the relevant personnel are only offset and not misplaced or in other situations, by adjusting the position of the treatment point, the two groups of means should still be equal. Then the determined means are equal, but they were not equal in the original assessment process. In this case, it is basically a case of wearing offset. It is only necessary to warn such personnel in the future and remind them of the correct way to wear the safety helmet. No on-site warning is required.

[0050] No matter how the processing points are adjusted, the corresponding related means between adjacent points after adjustment are not equal. First, the helmet may be misaligned, resulting in a large numerical deviation. When the helmet is misaligned, the arc curvature of each area is inconsistent, so the resulting associated lines will also change, resulting in unequal angles no matter how the adjustment is made. Or the relevant helmet is not worn, so no matter how the adjustment and processing are performed, the corresponding angle means cannot be equal.

[0051] Example 2

[0052] In the specific implementation process, compared with the above embodiment, this embodiment mainly identifies the numerical overlap between the associated angles during specific processing, and its specific execution end is the point angle processing end;

[0053] The point angle processing end, based on the characteristic contour determined in the image to be verified, identifies several groups of point lines determined within the characteristic contour, and starting from the edge points of the characteristic contour, sequentially determines the angle parameters between adjacent lines and determines the angle sequence. Then, by analyzing the proportion of the same angle in the angle sequence, it assesses whether the relevant personnel are wearing safety helmets. The specific method of performing the assessment is as follows:

[0054] Based on a number of feature points divided within the feature contour, a group of feature points at the edge of the feature contour are selected as initial points, a line connecting the initial point and its adjacent feature points is determined as a front line, a group of lines subsequent to the front line is then used as a back line, the intersection of the front line and the back line is used as an extension point, the front line is extended, and the angle between the extension line and the back line is locked as a first set of angles;

[0055] Then, the back line is used as the front line, and the back line of the previous line is determined, and the second set of angles is locked. Similarly, the angles between subsequent adjacent lines are confirmed one by one, and the confirmed angles are sorted according to the confirmation order to confirm the angle sequence;

[0056] Identify the angles with the same value that appear in the angle sequence. When the number of angles with the same value is ≥5, these angles with the same value will be calibrated as standard angles. Otherwise, no relevant calibration of standard angles will be performed.

[0057] Identify the total proportion of standard angles in the angle sequence, Zb, where Zb = the total number of standard angles ÷ the total number of angles in the angle sequence. If Zb ≥ Y1, a non-compliant wearing signal is generated through the signal generation end and directly displayed. Y1 is a preset value, and its specific value is determined by the operator based on experience. If Zb < Y1, a non-wearing signal is generated through the signal generation end and directly displayed, indicating that the relevant personnel are not wearing a safety helmet. The analyzed angle is abnormal, and the relevant personnel need to issue a relevant warning;

[0058] For example: If the initial point is A and the adjacent point is B, then line AB is the front line. Next, determine the set of lines following the front line as the back line. Assuming the next point is C, then line BC is the back line. Use the intersection of lines AB and BC as the extension point, extend line AB, and lock the angle between the extended line and line BC as the first set of angles.

[0059] Then, using line BC as the front line, continue to determine its back line CD and lock the second set of angles. Similarly, confirm the angles between subsequent adjacent lines one by one, and sort the confirmed angles according to the confirmation order to obtain the angle sequence.

[0060] Suppose that in this angle sequence, there are some angles with the same value. For example, there are multiple angles with the same value of 60 degrees. If the number of angles with the same value is ≥ 5, for example, there are 6 angles with the value of 60 degrees, then the angle of 60 degrees is calibrated as the standard angle.

[0061] Next, calculate the total percentage of standard angles in the angle sequence, Zb. Assuming there are 20 angles in the angle sequence and 8 standard angles (60-degree angles), then Zb = 8 ÷ 20 = 0.4.

[0062] If the preset value Y1 is 0.35, since Zb(0.4)≥Y1, a non-wearing signal is generated and displayed by the signal generating end. The words "Helmet wearing does not meet the standard" may be displayed on the display screen or a specific sound and light prompt may be emitted. If Zb<Y1, a non-wearing signal is generated and displayed, such as "Not wearing a helmet".

[0063] Some of the data in the above formulas are dimensionless and numerically calculated. Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0064] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. The construction site safety monitoring system based on drone vision is characterized by: include: The drone's visual system controls the drone's flight based on a preset trajectory and captures the construction scene in real time. The drone's internal visual system has its internal shooting angles pre-set before flight. The image primary processing end locks the relevant people appearing in the construction picture based on the set human features, and then extracts the associated images of the facial areas of the relevant people based on the preset face template. After adjusting the clarity of the associated images, the images to be verified are obtained and transmitted to the image primary analysis end; The initial image analysis end identifies the overall relevant contours of the image to be verified based on the processed image to be verified, and by eliminating the facial area in the image to be verified, determines whether the person associated with the image to be verified has worn any abnormal clothing. The specific method is as follows: Based on the facial region marked in the image to be verified, the center point of the facial region is locked, and a set of vertical lines passing through the center point and perpendicular to the horizontal base plane is constructed. The horizontal base plane is a preset horizontal plane. The intersection of the vertical lines and the overall outer contour of the image to be verified is locked, and the intersection point is marked as the processing point; The same plane where the center point and the point to be processed are located is marked as the selected surface, the part of the contour inside the selected surface is marked as the characteristic contour, and the characteristic contour is divided into several characteristic points at equal intervals; The points to be processed divide this feature contour into a left contour and a right contour, and prioritize identifying the associated angle between adjacent points in the left contour and the horizontal plane. Connect the adjacent feature points and translate the line until it intersects the horizontal plane. Lock the angle and mark it as J. i , where i represents the angle between the lines of different adjacent points, and the angles J determined by the left contour are i Perform mean processing to determine the left contour mean JZ1; Then process the right contour in the same way to determine the right contour mean JZ2; If JZ1≠JZ2, it means that the helmet is worn abnormally or the helmet is not worn, and subsequent relevant analysis and processing are required; Secondary adjustment recognition end, for the case where the person associated with the image to be verified has abnormal wearing, by making relevant adjustments to the processing points marked by the characteristic contour in the image to be verified, and identifying whether there is still abnormality after the relevant adjustment. If there is still abnormality, the point angle processing end is executed; The point angle processing end, based on the characteristic contour determined in the image to be verified, identifies several groups of point lines determined within this characteristic contour. Starting from the edge points of the characteristic contour, the angle parameters between adjacent lines are determined in sequence, and an angle sequence is determined. Then, by analyzing the proportion of the same angle in the angle sequence, it is assessed whether the relevant personnel are wearing safety helmets. The specific method is as follows: Based on a number of feature points divided within the feature contour, a group of feature points at the edge of the feature contour is selected as the initial point, the line connecting the initial point and its adjacent feature points is determined as the front line, and then a group of lines that appear successively after the front line is used as the back line. The intersection of the front line and the back line is used as the extension point to extend the front line, and the angle between the extension line and the back line is locked as the first group of angles; Then, the back line is used as the front line, and the back line of the previous line is determined, and the second set of angles is locked. Similarly, the angles between subsequent adjacent lines are confirmed one by one, and the confirmed angles are sorted according to the confirmation order to confirm the angle sequence; Identify the angles with the same value that appear in the angle sequence. When the number of angles with the same value is ≥5, these angles with the same value are calibrated as standard angles. Identify the total proportion of standard angles in the angle sequence Zb, where Zb = the total number of standard angles ÷ the total number of angles in the angle sequence. If Zb ≥ Y1, a wearing failure signal is generated through the signal generation end and directly displayed. Y1 is the preset value.

2. The construction site safety monitoring system based on drone vision according to claim 1 is characterized in that: If JZ1=JZ2, it means the helmet is worn normally and no further processing is required.

3. The construction site safety monitoring system based on drone vision according to claim 1 is characterized in that: The specific method of the secondary adjustment recognition end for recognizing whether the person associated with the image to be verified still has abnormal wearing is: Based on the points to be processed marked in the characteristic contour, the points to be processed are moved on the characteristic contour, and it is recorded whether the left contour mean JZ1 and the right contour mean JZ2 satisfy the following conditions in each different movement process: JZ1=JZ2: If so, a non-compliant wearing signal is generated through the signal generating end, and the relevant information is displayed directly.

4. The construction site safety monitoring system based on drone vision according to claim 3 is characterized in that: If the left contour mean JZ1 and the right contour mean JZ2 do not satisfy: JZ1=JZ2, it means that there is still an abnormality in the associated personnel, and the point angle processing end is executed for association analysis.

5. The construction site safety monitoring system based on drone vision according to claim 1 is characterized in that: If Zb<Y1, a no-wearing signal is generated through the signal generating end and displayed directly.

6. The construction site safety monitoring system based on drone vision according to claim 1 is characterized in that: When the number of angles with the same value is less than 5, no calibration of the standard angle is performed.

Images