CN114973684B

CN114973684B - Fixed-point monitoring method and system for construction site

Info

Publication number: CN114973684B
Application number: CN202210874108.8A
Authority: CN
Inventors: 杨翰翔; 赖晓俊
Original assignee: Shenzhen Lianhe Intelligent Technology Co ltd
Current assignee: Shenzhen Lianhe Intelligent Technology Co ltd
Priority date: 2022-07-25
Filing date: 2022-07-25
Publication date: 2022-10-14
Anticipated expiration: 2042-07-25
Also published as: CN114973684A

Abstract

The invention provides a fixed-point monitoring method and a fixed-point monitoring system for a building site, wherein each monitoring unmanned aerial vehicle is used for carrying out illegal behavior identification on each corresponding monitoring data stream by acquiring a monitoring data stream of an in-and-out transport vehicle of each fixed-point monitoring area in the building site, and based on a illegal behavior identification network associated with a building logic function partition corresponding to each fixed-point monitoring area, the illegal behavior identification information of each corresponding monitoring data stream is obtained, and early warning reminding is carried out on a patrol inspection duty terminal of each fixed-point monitoring area according to the illegal behavior identification information of each corresponding monitoring data stream. Therefore, fixed-point monitoring can be carried out by taking different building logic function partitions as independent monitoring units, and compared with global monitoring in the related art, the method has the advantages that the monitoring pertinence is better, and the monitoring early warning effect is improved.

Description

Construction site fixed-point monitoring method and system

Technical Field

The invention relates to the technical field of unmanned aerial vehicle monitoring, in particular to a construction site fixed-point monitoring method and system.

Background

For the construction site on site, the working environment is complex and changeable, safety accidents are often easy to occur, and how to implement real-time comprehensive management of the construction site is a very troublesome problem for relevant units. For example, in the case of a violation of passing in and out of a transportation vehicle, if a related violation is not discovered in time and an early warning is given, a serious safety accident may be caused. However, the current monitoring scheme is generally a global monitoring mode rather than a fixed-point unit monitoring mode, and the monitoring pertinence of the global monitoring mode is poor, and the final monitoring and early warning effect is also affected.

Disclosure of Invention

In order to overcome at least the above disadvantages of the prior art, the present invention provides a method and a system for site-specific monitoring of a construction site.

In a first aspect, the present invention provides a building site fixed-point monitoring method, which is applied to a fixed-point monitoring cloud platform, where the fixed-point monitoring cloud platform is in communication connection with multiple monitoring unmanned aerial vehicles, and the method includes:

acquiring a monitoring data stream of each monitoring unmanned aerial vehicle for the in-and-out transport vehicle of each fixed point monitoring area in the target building site;

carrying out violation behavior identification on each corresponding monitoring data stream based on the violation behavior identification network associated with the building logic function partition corresponding to each fixed point monitoring area to obtain violation behavior identification information of each corresponding monitoring data stream;

and early warning and reminding the inspection on-duty terminal of each fixed-point monitoring area according to the violation behavior identification information of each corresponding monitoring data stream.

In a second aspect, an embodiment of the present invention further provides a building site fixed-point monitoring system, where the target building site fixed-point monitoring system includes a fixed-point monitoring cloud platform and a plurality of monitoring unmanned aerial vehicles communicatively connected to the fixed-point monitoring cloud platform;

the fixed point monitoring cloud platform is used for:

acquiring a monitoring data flow of each monitoring unmanned aerial vehicle for the passing in and out of a transport vehicle of each fixed point monitoring area in the target building site;

and early warning and reminding the patrol inspection duty terminal in each fixed point monitoring area according to the violation behavior identification information of each corresponding monitoring data stream.

According to any one of the aspects, in the embodiment provided by the invention, the monitoring data stream of each monitoring unmanned aerial vehicle for the passing in and out of the transport vehicle in each fixed point monitoring area in the construction site is obtained, the violation behavior identification network associated with the building logic function partition corresponding to each fixed point monitoring area identifies the violation behavior of each corresponding monitoring data stream to obtain the violation behavior identification information of each corresponding monitoring data stream, and the inspection duty terminal of each fixed point monitoring area is early-warned according to the violation behavior identification information of each corresponding monitoring data stream. Therefore, fixed-point monitoring can be carried out by taking different building logic function partitions as independent monitoring units, and compared with global monitoring in the related art, the method has better monitoring pertinence and improves the monitoring and early warning effect.

Drawings

To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be called in the embodiments are briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention, and therefore should not be considered as limiting the scope, and those skilled in the art can also obtain other related drawings based on the drawings without creative efforts.

FIG. 1 is a schematic diagram of an application scenario of a construction site monitoring system according to an embodiment of the present invention;

FIG. 2 is a schematic flow chart of a construction site monitoring method according to an embodiment of the present invention;

fig. 3 is a schematic block diagram of a structure of a fixed-point monitoring cloud platform for implementing the above-described fixed-point monitoring method for a construction site according to an embodiment of the present invention.

Detailed Description

The present invention is described in detail below with reference to the drawings, and the specific operation methods in the method embodiments can also be applied to the apparatus embodiments or the system embodiments.

Fig. 1 is a schematic view of an application scenario of a construction site monitoring system 10 according to an embodiment of the present invention. The construction site monitoring system 10 may include a site monitoring cloud platform 100 and a monitoring drone 200 communicatively connected to the site monitoring cloud platform 100. The construction site monitoring system 10 shown in FIG. 1 is but one possible example, and in other possible embodiments, the construction site monitoring system 10 may include only at least some of the components shown in FIG. 1 or may include additional components.

In a possible design approach, the fixed point monitoring cloud platform 100 and the monitoring drone 200 in the construction site fixed point monitoring system 10 may cooperate to perform the construction site fixed point monitoring method described in the following method embodiments, and the steps performed by the fixed point monitoring cloud platform 100 and the monitoring drone 200 may be described in detail in the following method embodiments.

FIG. 2 is a schematic flow chart of a site-specific monitoring method for a construction site according to an embodiment of the present invention; in order to solve the technical problem in the background art, the construction site monitoring method provided by the present embodiment may be performed by the site monitoring cloud platform 100 shown in fig. 1. The construction site monitoring method will be described in detail below.

And step S110, acquiring a monitoring data stream of each monitoring unmanned aerial vehicle for the passing in and out of the transport vehicle of each fixed point monitoring area in the target building site.

For example, different fixed point monitoring areas can be designed for different construction sites, for example, a fixed point monitoring area 1, a fixed point monitoring area 2, a fixed point monitoring area 3, a fixed point monitoring area N and a fixed point monitoring area.

And step S120, carrying out violation behavior identification on each corresponding monitoring data stream based on the violation behavior identification network associated with the building logic function partition corresponding to each fixed point monitoring area, and obtaining violation behavior identification information of each corresponding monitoring data stream.

In this embodiment, the violation behavior recognition network may be obtained by pre-training, and a specific training process will be described in the following embodiments. The violation behavior identification network obtained through training can have the violation behavior identification capability, and further can carry out violation behavior identification on each corresponding monitoring data stream to obtain the violation behavior identification information of each corresponding monitoring data stream.

And step S130, according to the violation behavior identification information of each corresponding monitoring data stream, early warning and reminding are carried out on the inspection duty terminal of each fixed-point monitoring area.

For example, after the violation behavior identification information of each corresponding monitoring data stream is obtained, a corresponding early warning reminder can be sent out according to the inspection duty terminal of each fixed-point monitoring area bound in advance.

Based on the steps, the monitoring data streams of the transportation vehicles entering and exiting each fixed point monitoring area in the building site are acquired by each monitoring unmanned aerial vehicle, the violation behavior identification network associated with the building logic function partition corresponding to each fixed point monitoring area identifies the violation behavior of each corresponding monitoring data stream, the violation behavior identification information of each corresponding monitoring data stream is acquired, and the early warning and reminding are carried out on the inspection duty terminal of each fixed point monitoring area according to the violation behavior identification information of each corresponding monitoring data stream. Therefore, fixed-point monitoring can be carried out by taking different building logic function partitions as independent monitoring units, and compared with global monitoring in the related art, the method has better monitoring pertinence and improves the monitoring and early warning effect.

The step S120 can be implemented by the following steps in one design concept.

Step S121, acquiring a violation behavior identification network associated with the building logic function partition corresponding to each fixed point monitoring area;

and S122, extracting the target monitoring track characteristics of each corresponding monitoring data stream based on the violation behavior identification network associated with the building logic function partition corresponding to each fixed point monitoring area, identifying the violation behavior labels of the monitoring track characteristics, and acquiring the violation behavior identification information of each corresponding monitoring data stream.

In a possible design idea, based on the foregoing embodiment, an embodiment section of the violation behavior recognition network associated with the building logic function partition corresponding to each fixed-point monitoring area is described below, and the following detailed description is provided.

Step S101, obtaining training monitoring track characteristics corresponding to training monitoring data streams of building logic function partitions corresponding to each fixed-point monitoring area, inputting the training monitoring track characteristics to Q first-order AI training networks of a first AI training network (for example, a pre-trained AI training network), and obtaining Q illegal behavior identification information output by the Q first-order AI training networks.

In a design idea, before step S101, a training sample data set (e.g., a training sample data set with a characteristic component sequence) may be used to train to obtain an AI training network, which is not limited to this training process as long as the AI training network can be obtained, and for convenience of distinguishing, the AI training network is recorded as a first AI training network.

In one design approach, a first AI training network is used to predict Q types of violation labels, where Q may be a positive integer greater than or equal to 3. For example, if the first AI training network is used to classify overspeed violations, license plate non-specification violations, and unauthenticated vehicle violations, it means that the first AI training network is used to predict 3 types of violations labels, that is, Q is 3. If the first AI training network is used for classifying overspeed violation behaviors, license plate non-standard violation behaviors, unauthenticated vehicle violation behaviors and vehicle excessive distance violation behaviors, the first AI training network is used for predicting 4 violation behavior labels, namely Q is 4, and the rest can be done by analogy. For convenience of description, in the following embodiments, Q is taken as an example for explanation, and it is assumed that the first AI training network is used for classifying a violation label z, a violation label x, a violation label v, a violation label h, and a violation label o.

In one design approach, the first AI training network may include Q first-order AI training networks, a plurality of W first-order AI training networks, and a plurality of Q-1 first-order AI training networks, and W has a value of 2 to Q-2, i.e., the first AI training network may include a plurality of first-order AI training networks (e.g., Q first-order AI training networks), a plurality of second-order AI training networks, a plurality of third-order AI training networks, \ 823030, a plurality of Q-2 first-order AI training networks, and N/Q-1 first-order AI training networks (e.g., Q-1 first-order AI training networks).

When the component value sequence of Q is 5, the first AI training network may include a plurality of first-order AI training networks (e.g., 5 first-order AI training networks in total), a plurality of second-order AI training networks (e.g., 10 second-order AI training networks in total), a plurality of third-order AI training networks (e.g., 10 third-order AI training networks in total), and a plurality of fourth-order AI training networks (e.g., four fourth-order AI training networks in total).

In a design thought, Q first-order AI training networks are used for predicting Q types of violation behavior labels, and each first-order AI training network has a mapping relation with one type of violation behavior label, namely the Q first-order AI training networks correspond to the Q types of violation behavior labels one by one. Assuming that the first AI training network is used for classifying a violation behavior tag z, a violation behavior tag x, a violation behavior tag v, a violation behavior tag h and a violation behavior tag o, and the 5 first-order AI training networks are respectively a first-order AI training network N10, a first-order AI training network N11, a first-order AI training network N12, a first-order AI training network N13 and a first-order AI training network N14, then: the first-order AI training network N10 corresponds to the violation behavior label z, the first-order AI training network N11 corresponds to the violation behavior label x, the first-order AI training network N12 corresponds to the violation behavior label v, the first-order AI training network N13 corresponds to the violation behavior label h, and the first-order AI training network N14 corresponds to the violation behavior label o, on the basis that:

the violation identification information of the first-order AI training network N10 may be z or non-z (equivalent to xvho), where z denotes a violation label z, and non-z denotes not a violation label z. The violation identification information of the first-order AI training network N11 may be x or non-x (equivalent to zvho), where x denotes a violation label x and non-x denotes not a violation label x. The violation identification information of the first-order AI training network N12 may be v or non-v (equivalent to zxho), where v denotes a violation label v and non-v denotes not a violation label v. The violation identification information of the first-order AI training network N13 may be h or non-h (equivalent to zxvo), where h denotes a violation label h, and non-h denotes that the violation label h is not a violation label h. The violation identification information of the first-order AI training network N14 may be o or non-o (equivalent to zxvh), where o denotes the violation label o and non-o denotes not the violation label o.

By this design, the violation behavior recognition information of the 5 first-order AI training networks (first-order AI training network N10-first-order AI training network N14) is sequentially: z + xvho, x + zvho, v + zxho, h + zxvo, o + zxvh.

In a design idea, a mapping relation exists between each second-order AI training network in each second-order AI training network and two violation behavior labels, and the two violation behavior labels are used for predicting the two violation behavior labels, that is, one second-order AI training network corresponds to the two violation behavior labels, and the two violation behavior labels corresponding to different second-order AI training networks are not identical.

Assuming that Q is 5 and 10 second-order AI training networks are second-order AI training network N20 — second-order AI training network N29, then: the second-order AI training network N20 corresponds to the violation label z and the violation label x, the second-order AI training network N21 corresponds to the violation label z and the violation label v, the second-order AI training network N22 corresponds to the violation label z and the violation label h, the second-order AI training network N23 corresponds to the violation label z and the violation label o, the second-order AI training network N24 corresponds to the violation label x and the violation label v, and so on.

The violation identification information of the second-order AI training network N20 is z, or x, or non-zx (equivalent to vho), where z represents a violation label z, x represents a violation label x, and non-zx represents neither the violation label z nor the violation label x. The violation identification information of the second-order AI training network N21 may be z, or v, or non-zv (equivalent to xho), and so on. Obviously, the violation identification information of the 10 second-order AI training networks (second-order AI training network N20 — second-order AI training network N29) is in turn: z + x + vho, z + v + xho, z + h + xvo, z + o + xvh, x + v + zhoo, x + h + zvo, x + o + zvh, v + h + zxo, v + o + zxh, h + o + zxv.

In a design idea, a mapping relation exists between each three-order AI training network in each three-order AI training network and three violation behavior labels, and the mapping relation is used for predicting the three violation behavior labels, namely, one three-order AI training network corresponds to the three violation behavior labels, and the three violation behavior labels corresponding to different three-order AI training networks are not identical.

Suppose Q is 5, 10 three-order AI training networks are three-order AI training networks N30-N39, the three-order AI training network N30 corresponds to the violation behavior label z, the violation behavior label x and the violation behavior label v, the three-order AI training network N31 corresponds to the violation behavior label z, the violation behavior label x and the violation behavior label h, and so on. The violation identification information of the third-order AI training network N30 is z, or x, or v, or non-zxv (equal to ho), wherein z represents a violation label z, x represents a violation label x, v represents a violation label v, and non-zxv represents that the violation label z is not a violation label x, nor a violation label v, and so on. Obviously, the violation behavior recognition information of the above 10 third-order AI training networks (third-order AI training network N30-third-order AI training network N39) is in turn: z + x + v + ho, z + x + h + vo, z + x + o + vh, z + v + h + xo, z + v + o + xh, z + h + o + xv, x + v + h + zo, x + v + o + zh, x + h + o + zv, v + h + o + zx.

According to the design, a mapping relation exists between each W-order (W value is 2-Q-2) AI training network and W violation behavior labels, and the W-order AI training networks are used for predicting the W violation behavior labels, namely, one W-order AI training network corresponds to the W violation behavior labels, and the W violation behavior labels corresponding to different W-order AI training networks are not identical. Obviously, when Q is 5, the component value sequence range of W may be 2, 3.

In a design idea, a mapping relation exists between a plurality of Q-1-order AI training networks and Q violation behavior labels, and the Q violation behavior labels are used for predicting the Q violation behavior labels, namely the Q violation behavior labels are corresponding to the Q-1-order AI training networks.

Assuming that Q is 5, there are multiple () fourth-order AI training networks, i.e., a fourth-order AI training network N40, and the fourth-order AI training network N40 corresponds to the violation labels z, x, v, h. The fourth-order AI training network N40 also corresponds to the violation labels z, x, v, o. The fourth-order AI training network N40 also corresponds to the violation behavior labels z, x, h, o. The fourth-order AI training network N40 also corresponds to the violation behavior labels z, v, h, o. The fourth-order AI training network N40 also corresponds to the violation behavior labels x, v, h, o. The violation identification information of the fourth-order AI training network N40 is z, or x, or v, or h, or o (equivalent to non-zxvh), that is, the violation identification information of the fourth-order AI training network N40 is: z + x + v + h + o.

Based on the first AI training network, for step S101, after obtaining the training monitoring trajectory feature corresponding to the training monitoring data stream, the training monitoring trajectory feature may be input to Q first-order AI training networks of the first AI training network, such as the first-order AI training network N10-the first-order AI training network N14. Because the first-order AI training networks are used for predicting the characteristics of the training monitoring track, Q first-order AI training networks can output Q illegal behavior identification information, namely, each first-order AI training network can output one illegal behavior identification information. For example, the violation identification information output by the first-order AI training network N10 is z or non-z, the violation identification information output by the first-order AI training network N11 is x or non-x, the violation identification information output by the first-order AI training network N12 is v or non-v, the violation identification information output by the first-order AI training network N13 is h or non-h, and the violation identification information output by the first-order AI training network N14 is o or non-o.

In a design idea, due to the characteristics of the building logic function partition corresponding to each fixed-point monitoring area, a training monitoring data stream for the building logic function partition corresponding to each fixed-point monitoring area can be obtained, after the training monitoring data stream is obtained, a monitoring track characteristic corresponding to the training monitoring data stream can be determined, and in order to distinguish conveniently, a monitoring track characteristic needing to be input to a first AI training network can be recorded as a training monitoring track characteristic.

In a design idea, in order to obtain a training monitoring trajectory feature corresponding to a training monitoring data stream of a building logic function partition corresponding to each fixed point monitoring area, the training monitoring trajectory feature may be obtained by the following steps:

step S1011, acquiring a training monitoring data stream of a building logic function partition corresponding to each fixed point monitoring area, and extracting a plurality of monitoring characteristic configuration information aiming at the building logic function partition corresponding to each fixed point monitoring area to cluster a plurality of key track sections matched with current training monitoring data stream nodes at a first violation monitoring time period of the training monitoring data stream, wherein each key track section cluster comprises a track characteristic field of the current training monitoring data stream node aiming at the monitoring characteristic configuration information;

step S1012, determining a current key track segment characteristic set required by the current training monitoring data stream node from the building logic function partition according to the plurality of key track segment clusters;

step S1013, when determining a plurality of violation monitoring time sequence segments aiming at current key track segment characteristic sets respectively needed by the current training monitoring data stream node, generating a candidate key track segment clustering sequence of the current training monitoring data stream node corresponding to the building logic function partition by using key track segment contents respectively contained by the current key track segment characteristic sets, wherein the plurality of violation monitoring time sequence segments comprise the first violation monitoring time sequence segment and a violation monitoring time sequence segment behind the first violation monitoring time sequence segment;

step S1014, when acquiring a plurality of candidate key track segment clustering sequences generated by a reference violation monitoring time sequence corresponding to the building logic function partition, screening the plurality of candidate key track segment clustering sequences to obtain a target key track segment clustering sequence;

step S1015, performing feature extraction on the target key trajectory segment clustering sequence to obtain the training monitoring trajectory features.

For example, the determining a current key trajectory segment feature set required by the current training monitoring data stream node from the building logical function partition according to the plurality of key trajectory segment clusters includes:

classifying a plurality of key trajectory segment matching templates in the building logical function partition;

sequentially determining each key track segment matching template as a current track segment matching template, and executing the following steps until all key track segment matching templates are traversed:

calculating the matching degree of the current track segment matching template and each training monitoring data stream node, wherein the matching degree is determined according to the track floating feature aimed at by the current track segment matching template, the track floating feature included by the training monitoring data stream node and the clustering of the plurality of key track segments;

performing weight fusion calculation on all the matching degrees of the current track segment matching template to generate a candidate matching degree between each current training monitoring data stream node and the current track segment matching template;

determining the matched key track segment characteristics corresponding to the key track segment matching template corresponding to the maximum candidate matching degree in all the candidate matching degrees as the current key track segment characteristic set of the current training monitoring data stream node in the first violation monitoring time period;

for example, the calculating the matching degree between the current track segment matching template and each of the training monitoring data stream nodes includes:

sequentially determining each training monitoring data stream node as a current training monitoring data stream node, and executing the following steps until all the training monitoring data stream nodes are traversed:

when the key track segment cluster required by the track floating feature of the current training monitoring data stream node and the current track segment matching template is the key track segment cluster corresponding to the current training monitoring data stream node, determining the matching degree of the current track segment matching template and the current training monitoring data stream node as a first preset matching degree;

and when the key track segment cluster required by the current track segment matching template and the track floating feature of the current training monitoring data stream node is not the key track segment cluster corresponding to the current training monitoring data stream node, determining the matching degree to be a second preset matching degree.

For example, when the multiple candidate key track segment clustering sequences generated by the reference violation monitoring time sequence corresponding to the building logic function partition are obtained, the screening of the multiple candidate key track segment clustering sequences to obtain a target key track segment clustering sequence includes:

dividing the candidate key track segment cluster sequences into multiple candidate key track segment cluster sequences;

determining a target key track segment clustering sequence for each type of the candidate key track segment clustering sequences;

for example, the determining a target key track segment clustering sequence for each type of the candidate key track segment clustering sequences includes:

after one class of candidate key track segment clustering sequences is obtained, determining one candidate key track segment clustering sequence in the class of candidate key track segment clustering sequences as a current candidate key track segment clustering sequence, determining the other candidate key track segment clustering sequence as a first candidate key track segment clustering sequence, and executing the following steps until all candidate key track segment clustering sequences in the class of candidate key track segment clustering sequences are traversed and ended:

obtaining tag range values of every two corresponding clustering tags in the current candidate key track segment clustering sequence and the first candidate key track segment clustering sequence, determining the candidate key track segment clustering sequence corresponding to the tag range values as a new current candidate key track segment clustering sequence, and determining one candidate key track segment clustering sequence in the candidate key track segment clustering sequences remaining in one class of candidate key track segment clustering sequences as the first candidate key track segment clustering sequence;

and after the traversal is completed, determining a finally determined current candidate key track segment clustering sequence as the target key track segment clustering sequence of the candidate key track segment clustering sequences.

By means of the design, in step S101, Q pieces of violation behavior identification information output by Q first-order AI training networks can be obtained.

Step S102, if Q pieces of illegal behavior identification information correspond to R kinds of illegal behavior labels, and R is larger than 1 and smaller than Q, a second AI training network is selected from all R-order AI training networks of the first AI training network, and the second AI training network and the R kinds of illegal behavior labels have a mapping relation; and inputting the training monitoring track characteristics to the second AI training network to obtain violation behavior identification information output by the second AI training network, and determining a target violation behavior label corresponding to the training monitoring track characteristics based on the violation behavior identification information. Or if the Q pieces of violation behavior identification information correspond to a violation behavior label, determining the violation behavior label as a target violation behavior label corresponding to the training monitoring track characteristic. Or if the Q illegal behavior identification information corresponds to the Q illegal behavior labels, selecting the Q-1 order AI training network of the first AI training network as a second AI training network; and inputting the training monitoring track characteristics to the second AI training network to obtain violation behavior identification information output by the second AI training network, and determining a target violation behavior label corresponding to the training monitoring track characteristics based on the violation behavior identification information.

For example, when Q first-order AI training networks output Q (e.g., 5) pieces of violation identification information, the Q pieces of violation identification information may be z or non-z, x or non-x, v or non-v, h or non-h, o or non-o.

If Q pieces of violation behavior identification information are z, non-x, non-v, non-h and non-o, the Q pieces of violation behavior identification information correspond to the same violation behavior label z, so that the violation behavior label z can be determined as a target violation behavior label corresponding to the training monitoring track characteristic. For the case that the same violation behavior label is x, v, h, o, similar to the case that the same violation behavior label is z, the description is not repeated here.

If the Q pieces of illegal behavior identification information are z, x, non-v, non-h and non-o, the Q pieces of illegal behavior identification information correspond to two illegal behavior labels, namely R is 2, and the two illegal behavior labels are z and x, so that a second AI training network needs to be selected from all second-order AI training networks of the first AI training network, and the second AI training network corresponds to the illegal behavior label z and the illegal behavior label x. Referring to the above embodiment, the second-order AI training network N20 corresponds to the violation behavior tag z and the violation behavior tag x, and thus the second AI training network is the second-order AI training network N20.

For the case that the two violation behavior labels are z and v, z and h, and the like, the case is similar to the case that the two violation behavior labels are z and x, except that the second AI training network is different, and repeated description is omitted here.

After knowing that the second AI training network is the second-order AI training network N20, the training monitoring trajectory features may be input to the second-order AI training network N20, and the second-order AI training network N20 may output a plurality of violation behavior recognition information, which may be z, or x, or non-zx. On the basis, if the violation identification information output by the second-order AI training network N20 is z, it is determined that the target violation label corresponding to the training monitoring track feature is the violation label z based on the violation identification information. If the violation behavior identification information output by the second-order AI training network N20 is x, determining that a target violation behavior label corresponding to the training monitoring track characteristic is a violation behavior label x based on the violation behavior identification information. If the violation identification information output by the second-order AI training network N20 is non-zx, it indicates that the target violation label corresponding to the training monitoring track feature cannot be determined yet.

Assuming that the Q pieces of violation identification information are z, x, v, non-h, and non-o, the Q pieces of violation identification information correspond to three violation labels, that is, R is 3, and the three violation labels are z, x, and v, so that a second AI training network needs to be selected from all three-order AI training networks of the first AI training network, and the second AI training network corresponds to the violation label z, the violation label x, and the violation label v. Referring to the above embodiment, the third-order AI training network N30 corresponds to the violation label z, the violation label x, and the violation label v, and thus the second AI training network is the third-order AI training network N30. For the case that the three violation behavior labels are zxh, zxo and the like, the description is not repeated here, which is similar to the case that the three violation behavior labels are zxv.

After knowing that the second AI training network is a third-order AI training network N30, the training monitoring trajectory features may be input to the third-order AI training network N30, and the third-order AI training network N30 may output a plurality of violation behavior recognition information, i.e., z, or x, or v, or non-zxv. And if the illegal behavior identification information is z, determining that the target illegal behavior label is the illegal behavior label z based on the illegal behavior identification information. And if the illegal behavior identification information is x, determining that the target illegal behavior label is the illegal behavior label x based on the illegal behavior identification information. And if the violation behavior identification information is v, determining that the target violation behavior label is the violation behavior label v based on the violation behavior identification information. And if the violation identification information is non-zxv, the target violation label cannot be determined.

Assuming that the Q pieces of violation identification information are z, x, v, h and non-o, the Q pieces of violation identification information correspond to four violation labels, that is, R is 4, and the four violation labels are z, x, v and h, so that a second AI training network needs to be selected from all four-stage AI training networks of the first AI training network, and the second AI training network corresponds to the violation labels z, x, v and h. Since there is only one fourth-order AI training network N40, and the fourth-order AI training network N40 corresponds to the violation behavior labels z, x, v, and h, the second AI training network may be the fourth-order AI training network N40. For the cases that the four violation behavior labels are zxvo, zxho, and the like, similar to the cases that the four violation behavior labels are zxvh, the second AI training networks are all the fourth-order AI training network N40, and the details are not repeated here.

After the second AI training network is the fourth-order AI training network N40, the training monitoring track characteristics are input to the fourth-order AI training network N40, and the fourth-order AI training network N40 outputs a plurality of violation behavior recognition information, i.e., z, or x, or v, or h, or o. And if the violation behavior identification information is z, determining that the target violation behavior label is a violation behavior label z based on the violation behavior identification information. And if the illegal behavior identification information is x, determining that the target illegal behavior label is the illegal behavior label x based on the illegal behavior identification information. And if the violation behavior identification information is v, determining that the target violation behavior label is the violation behavior label v based on the violation behavior identification information. And if the violation behavior identification information is h, determining that the target violation behavior label is the violation behavior label h based on the violation behavior identification information. And if the illegal behavior identification information is o, determining that the target illegal behavior label is the illegal behavior label o based on the illegal behavior identification information.

If Q pieces of violation behavior recognition information are z, x, v, h, and o, the Q pieces of violation behavior recognition information correspond to Q (for example, five) kinds of violation behavior labels, and therefore, a Q-1-order AI training network (for example, a fourth-order AI training network) of the first AI training network may be selected as the second AI training network, that is, the second AI training network may be the fourth-order AI training network N40.

After knowing that the second AI training network is the fourth-order AI training network N40, the training monitoring trajectory features are input to the fourth-order AI training network N40, and the fourth-order AI training network N40 outputs a plurality of violation behavior recognition information, i.e., z, or x, or v, or h, or o. And if the illegal behavior identification information is z, determining that the target illegal behavior label is the illegal behavior label z based on the illegal behavior identification information. And if the violation behavior identification information is x, determining that the target violation behavior tag is the violation behavior tag x based on the violation behavior identification information. And if the violation behavior identification information is v, determining that the target violation behavior label is the violation behavior label v based on the violation behavior identification information. And if the violation behavior identification information is h, determining that the target violation behavior label is the violation behavior label h based on the violation behavior identification information. And if the illegal behavior identification information is o, determining that the target illegal behavior label is the illegal behavior label o based on the illegal behavior identification information.

Therefore, in step S102, a target violation behavior label corresponding to the training monitoring track feature may be determined (for convenience of distinguishing, the violation behavior label corresponding to the training monitoring track feature is recorded as the target violation behavior label).

And S103, determining a characteristic component sequence of the training monitoring track characteristic based on the target violation behavior label.

For example, if the target violation label is violation label z, the characteristic component sequence is a first component value sequence, and the first component value sequence represents violation label z. And if the target violation behavior label is a violation behavior label x, the characteristic component sequence is a second component value sequence, and the second component value sequence represents the violation behavior label x. If the target violation behavior label is the violation behavior label v, the characteristic component sequence is a third component value sequence, and the third component value sequence represents the violation behavior label v. If the target violation behavior label is a violation behavior label h, the characteristic component sequence is a fourth component value sequence, and the fourth component value sequence represents the violation behavior label h. And if the target violation behavior label is a violation behavior label o, the characteristic component sequence is a fifth component value sequence, and the fifth component value sequence represents the violation behavior label o.

Step S104, performing network optimization on the first AI training network based on the training monitoring track feature and the feature component sequence of the training monitoring track feature, and determining a second AI training network based on the network-optimized AI training network, where the second AI training network is used to determine a violation behavior label corresponding to the monitoring data stream to be identified, and the process may refer to subsequent embodiments.

In a design idea, after obtaining a training monitoring trajectory feature and a feature component sequence of the training monitoring trajectory feature, the training monitoring trajectory feature and the feature component sequence may be used as a training sample data set, that is, a training sample data set with a feature component sequence is obtained, and then, network optimization may be performed on a first AI training network based on the training sample data set, for example, parameters of all AI training networks (such as a first-order AI training network, a second-order AI training network, and the like) in the first AI training network are optimized, and a training process of the first AI training network is not limited. After the network optimization is performed on the first AI training network, the AI training network after the network optimization can be obtained, and the second AI training network is determined based on the AI training network after the network optimization.

In a design idea, for step S101, when acquiring training monitoring trajectory features corresponding to training monitoring data streams of building logic function partitions corresponding to each fixed-point monitoring area, a training sample data set may be acquired first, where the training sample data set may include monitoring trajectory features corresponding to training monitoring data streams of building logic function partitions corresponding to each fixed-point monitoring area. Based on this, one monitoring trajectory feature may be traversed from the training sample data set as a training monitoring trajectory feature. On this basis, for step S104, when determining the second AI training network based on the network-optimized AI training network, it may be determined whether the network-optimized AI training network satisfies a training termination condition; if not, the network-optimized AI training network may be determined as the first AI training network, another monitoring trajectory feature is traversed from the training sample data set as a training monitoring trajectory feature, and the operation of inputting the training monitoring trajectory feature to Q first-order AI training networks of the first AI training network is returned (for example, the operation returns to step S101). If so, a second AI training network may be determined based on the network optimized AI training network.

In a design idea, after a first AI training network is subjected to network optimization to obtain a network-optimized AI training network, a second AI training network is determined based on the network-optimized AI training network, which may include but is not limited to the following cases:

and in the first situation, determining the network-optimized AI training network as a second AI training network.

For example, the first AI training network includes Q first-order AI training networks, a plurality of W-order AI training networks, and a plurality of Q-1-order AI training networks, and W has a value of 2 to Q-2, based on which, after network optimization of the first AI training network, the network-optimized AI training network also includes Q first-order AI training networks, a plurality of W-order AI training networks, and a plurality of Q-1-order AI training networks, and when the network-optimized AI training network is determined as the second AI training network, the second AI training network also includes Q first-order AI training networks, a plurality of W-order AI training networks, and a plurality of Q-1-order AI training networks.

And in the second situation, selecting a Q-1 order AI training network from the network-optimized AI training networks, and determining a second AI training network based on the Q-1 order AI training network, wherein the second AI training network at least comprises the Q-1 order AI training network.

For example, the first AI training network may include Q first-order AI training networks, a plurality of W-order AI training networks, and a plurality of Q-1-order AI training networks, and the network-optimized AI training network may also include Q first-order AI training networks, a plurality of W-order AI training networks, and a plurality of Q-1-order AI training networks, based on which the (a total of) Q-1-order AI training networks may be selected from the network-optimized AI training networks, and thus, the second AI training network may include the Q-1-order AI training network.

After the second AI training network is obtained, the violation behavior tag corresponding to the monitoring data stream to be identified may be determined based on the second AI training network, and the process is a detection process, which may refer to the subsequent embodiments.

According to the technical scheme, a large number of training monitoring data streams can be obtained, namely the data streams of the building logic function partitions corresponding to each fixed-point monitoring area are all the training monitoring data streams, so that a large number of training sample data sets are used for carrying out network optimization on the AI training network, the accurate and reliable AI training network is trained, the AI training network can accurately identify the violation behavior labels, and subsequent early warning reminding is carried out based on the violation behavior labels. In order to add a characteristic component sequence to a training monitoring data stream, a first AI training network can be trained by using a small amount of training sample data sets with accurate characteristic component sequences, a violation behavior label of the training monitoring data stream is determined by using the first AI training network, and the characteristic component sequence is added to the training monitoring data stream based on the violation behavior label, so that the automatic addition of the characteristic component sequence is realized, the characteristic component sequence of the training monitoring data stream does not need to be manually marked, and the training efficiency is improved.

During the training process, the method may comprise:

step S201, a training sample data set is obtained, where the training sample data set includes a training sample data set with a feature component sequence, that is, each training sample data set in the training sample data set includes a monitoring trajectory feature and a feature component sequence of a training monitoring data stream, and the obtaining manner of the training sample data set is not limited. And training based on the training sample data set to obtain a first AI training network, wherein the first AI training network comprises Q first-order AI training networks, a plurality of W-order AI training networks and a plurality of Q-1-order AI training networks.

Step S202, a training sample data set is obtained, where the training sample data set includes monitoring trajectory features corresponding to the training monitoring data stream of the building logic function partition corresponding to each fixed-point monitoring area, that is, the training sample data set may include a plurality of monitoring trajectory features without feature component sequences.

Step S203, traversing a monitoring trajectory feature from the training sample data set as a training monitoring trajectory feature.

For example, all the monitored trajectory features in the training sample data set may be sorted, and are not particularly limited. In step S203, the first monitoring trajectory feature in the training sample data set is used as the training monitoring trajectory feature.

Step S204, aiming at the characteristics of the currently traversed training monitoring track, inputting the characteristics of the training monitoring track into Q first-order AI training networks of the first-order AI training network to obtain Q illegal behavior identification information output by the Q first-order AI training networks.

And S205, determining a target violation behavior label corresponding to the training monitoring track characteristic based on the Q pieces of violation behavior identification information, wherein the specific determination mode refers to the step S102, and repeated description is omitted here. And if the target violation behavior label is determined based on the Q violation behavior identification information, deleting the training monitoring track characteristics from the training sample data set, and executing the step S206. And if the target illegal behavior label is not determined based on the Q pieces of illegal behavior identification information, deleting the training monitoring track characteristic from the training sample data set, or moving the training monitoring track characteristic to the tail of the training sample data set to form the last monitoring track characteristic.

And S206, determining a characteristic component sequence of the training monitoring track characteristic based on the target violation behavior label.

Step S207, forming a training sample data set from the training monitoring trajectory feature and the feature component sequence, and adding the training sample data set to the training sample data set, that is, adding a training sample data set with a feature component sequence to the training sample data set.

Step S208, performing network optimization on the first AI training network based on the training sample data set (for example, performing optimization on parameters of each AI training network in the first AI training network), to obtain an AI training network after network optimization.

Obviously, since the training sample data set already includes the training sample data set composed of the training monitor trajectory feature and the feature component sequence, the network optimization may be performed on the first AI training network based on the training monitor trajectory feature and the feature component sequence.

Step S209 determines whether the network-optimized AI training network satisfies a training termination condition.

If not, step S210 may be performed, and if yes, step S211 may be performed.

For example, whether the monitoring trajectory feature exists in the training sample data set or not may be determined, if so, it is determined that the AI training network after network optimization does not satisfy the training termination condition, and if not, it is determined that the AI training network after network optimization satisfies the training termination condition.

For another example, it may be determined whether the iterative training number of the AI training network (each time the first AI training network is network optimized, the iterative training number may be increased by 1) reaches a number threshold (configured according to experience), and if so, it may be determined that the network-optimized AI training network satisfies a training termination condition, and if not, it may be determined that the network-optimized AI training network does not satisfy the training termination condition.

For another example, it may be determined whether a training duration of the AI training network (from a time when network optimization of the AI training network is started to a current time) reaches a duration threshold (configured empirically), and if so, it is determined that the network-optimized AI training network satisfies a training termination condition, and if not, it is determined that the network-optimized AI training network does not satisfy the training termination condition.

Of course, the above manners are only examples of determining whether the AI training network satisfies the training termination condition, and the determination manner is not limited as long as it can be determined whether the AI training network satisfies the training termination condition.

Step S210, determining the network-optimized AI training network as a first AI training network, traversing another monitoring trajectory feature (such as the first monitoring trajectory feature) from the training sample data set as a training monitoring trajectory feature, and returning to execute step S204.

Step S211, determining a second AI training network based on the network optimized AI training network. For example, the network-optimized AI training network is determined as the second AI training network. Or, a Q-1 order AI training network is selected from the network optimized AI training networks, and the second AI training network at least comprises the Q-1 order AI training network.

To this end, a second AI training network may be available, which is capable of performing a detection process based on the second AI training network.

In the application process, that is to say for step S120, the following steps may be included.

Step S301, obtaining target monitoring track characteristics corresponding to the monitoring data stream to be identified.

In a design idea, a data stream for an internet of things device may be a normal data stream or a training monitoring data stream, and when the data stream for the internet of things device is the training monitoring data stream, a violation behavior tag corresponding to the data stream needs to be detected, so that the data stream is recorded as a monitoring data stream to be identified. And aiming at the data streams of the building logic function partitions corresponding to each fixed point monitoring area, the data streams are training monitoring data streams, and violation behavior labels corresponding to the data streams need to be detected, so that the data streams are recorded as the monitoring data streams to be identified.

After the monitoring data stream to be identified is obtained, data information is extracted from the monitoring data stream to be identified, the data information at least comprises packet header information and/or load information of the monitoring data stream to be identified, the data information is input into a trained self-encoder model, and monitoring track characteristics, namely target monitoring track characteristics, which are output by the self-encoder model and correspond to the monitoring data stream to be identified are obtained.

The implementation process of step S301 can refer to step S1011 to step S1013, and will not be described herein again.

Step S302, inputting the target monitoring track characteristic to a second AI training network to obtain violation behavior identification information output by the second AI training network, and determining a target violation behavior label corresponding to the target monitoring track characteristic, namely, a target violation behavior label corresponding to the monitoring data stream to be identified based on the violation behavior identification information.

In a design idea, the second AI training network may include a plurality of Q-1 order AI training networks, and based on this, after obtaining a target monitoring trajectory feature corresponding to a monitoring data stream to be identified, the target monitoring trajectory feature may be input to the Q-1 order AI training network of the second AI training network, so as to obtain violation behavior identification information output by the Q-1 order AI training network, and a target violation behavior tag corresponding to the target monitoring trajectory feature may be determined based on the violation behavior identification information.

For example, taking the Q-1 order AI training network as the fourth order AI training network N40 as an example, the target monitoring trajectory feature may be input to the fourth order AI training network N40, and the fourth order AI training network N40 outputs a plurality of violation behavior identification information, i.e., z, or x, or v, or h, or o. And if the violation behavior identification information is z, determining that the target violation behavior label corresponding to the target monitoring track characteristic is a violation behavior label z based on the violation behavior identification information. And if the violation behavior identification information is x, determining that the target violation behavior label corresponding to the target monitoring track characteristic is the violation behavior label x based on the violation behavior identification information, and so on.

In another design idea, the second AI training network may include Q first-order AI training networks, a plurality of W-order (W is a value of 2 to Q-2) AI training networks, and a plurality of Q-1-order AI training networks, and based on this, after obtaining the target monitoring trajectory feature corresponding to the monitoring data stream to be identified, the target monitoring trajectory feature may be input to the Q first-order AI training networks of the second AI training network, so as to obtain Q violation behavior identification information output by the Q first-order AI training networks.

If the Q pieces of violation behavior identification information correspond to one violation behavior label, the violation behavior label can be determined as a target violation behavior label corresponding to the target monitoring track characteristic. If the Q pieces of illegal behavior identification information correspond to R (R is more than 1 and less than Q) illegal behavior labels, a second AI training network can be selected from all R-order AI training networks of the second AI training network, the target monitoring track characteristics are input to the second AI training network, illegal behavior identification information output by the second AI training network is obtained, and the target illegal behavior labels corresponding to the target monitoring track characteristics are determined based on the illegal behavior identification information. And if the Q pieces of illegal behavior identification information correspond to the Q kinds of illegal behavior labels, selecting the Q-1 order AI training network as a second AI training network, inputting the target monitoring track characteristics to the second AI training network to obtain illegal behavior identification information output by the second AI training network, and determining the target illegal behavior label corresponding to the target monitoring track characteristics based on the illegal behavior identification information.

For example, if Q pieces of violation identification information are z, non-x, non-v, non-h, and non-o, the Q pieces of violation identification information correspond to the same violation label z, and therefore, the target violation label corresponding to the target monitoring track feature is the violation label z. If the Q pieces of violation behavior identification information are z, x, non-v, non-h and non-o, the Q pieces of violation behavior identification information correspond to two violation behavior labels, namely z and x, so that a second AI training network, namely a second-order AI training network N20, is selected from all second-order AI training networks of the second AI training network. The target monitoring track characteristics are input to the second-order AI training network N20, and the second-order AI training network N20 outputs a plurality of violation behavior identification information. And if the violation behavior identification information is z, the target violation behavior label corresponding to the target monitoring track characteristic is a violation behavior label z. And if the violation behavior identification information is x, the target violation behavior label corresponding to the target monitoring track characteristic is the violation behavior label x.

Assuming that the Q illegal behavior identification information is z, x, v, non-h and non-o, the Q illegal behavior identification information corresponds to three illegal behavior labels, namely z, x and v, so that a second AI training network, namely a third-order AI training network N30 is selected from all third-order AI training networks of the second AI training network. And then, inputting the target monitoring track characteristics to a three-order AI training network N30, wherein the three-order AI training network N30 outputs a plurality of violation behavior identification information, and if the violation behavior identification information is z, the target violation behavior label corresponding to the target monitoring track characteristics is a violation behavior label z. And if the violation behavior identification information is x, the target violation behavior label corresponding to the target monitoring track characteristic is the violation behavior label x. And if the violation behavior identification information is v, the target violation behavior label corresponding to the target monitoring track characteristic is a violation behavior label v.

Assuming that the Q pieces of violation identification information are z, x, v, h, and non-o, the Q pieces of violation identification information correspond to four violation labels, namely z, x, v, and h, and therefore, one second AI training network, namely a fourth-order AI training network N40, is selected from all the fourth-order AI training networks of the second AI training network. And then, inputting the target monitoring track characteristics to a three-order AI training network N30, wherein the three-order AI training network N30 outputs a plurality of violation behavior identification information, and if the violation behavior identification information is z, the target violation behavior label corresponding to the target monitoring track characteristics is a violation behavior label z. And if the violation behavior identification information is x, the target violation behavior label corresponding to the target monitoring track characteristic is the violation behavior label x. And if the violation behavior identification information is v, the target violation behavior label corresponding to the target monitoring track characteristic is a violation behavior label v. And if the violation behavior identification information is h, the target violation behavior label corresponding to the target monitoring track characteristic is a violation behavior label h. And if the illegal behavior identification information is o, the target illegal behavior label corresponding to the target monitoring track characteristic is the illegal behavior label o.

Assuming that the Q illegal behavior recognition information are z, x, v, h and o, the Q illegal behavior recognition information correspond to the Q illegal behavior labels, and therefore, the Q-1-order AI training network (for example, the fourth-order AI training network) of the second AI training network is selected as the second AI training network, that is, the second AI training network is the fourth-order AI training network N40. And then, inputting the target monitoring track characteristics to a fourth-order AI training network N40, wherein the fourth-order AI training network N40 outputs a plurality of violation behavior identification information, and if the violation behavior identification information is z, the target violation behavior label corresponding to the target monitoring track characteristics is a violation behavior label z. And if the violation behavior identification information is x, the target violation behavior label corresponding to the target monitoring track characteristic is the violation behavior label x. And if the violation behavior identification information is v, the target violation behavior label corresponding to the target monitoring track characteristic is a violation behavior label v. And if the violation behavior identification information is h, the target violation behavior label corresponding to the target monitoring track characteristic is a violation behavior label h. And if the violation behavior identification information is o, the target violation behavior label corresponding to the target monitoring track characteristic is a violation behavior label o.

Fig. 3 illustrates a hardware structure of the fixed-point monitoring cloud platform 100 for implementing the construction site fixed-point monitoring method, according to an embodiment of the present invention, and as shown in fig. 3, the fixed-point monitoring cloud platform 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

In a specific implementation process, the plurality of processors 110 execute the computer-executable instructions stored in the machine-readable storage medium 120, so that the processors 110 may execute the construction site fixed-point monitoring method according to the above method embodiment, the processors 110, the machine-readable storage medium 120, and the communication unit 140 are connected through the bus 130, and the processors 110 may be configured to control the transceiving action of the communication unit 140, so as to perform data transceiving with the monitoring drone 200.

Machine-readable storage medium 120 may store data and/or instructions. In some embodiments, machine-readable storage medium 120 may store data and/or instructions for fixed point monitoring cloud platform 100 to perform or use to perform the exemplary methods described in this disclosure. In some embodiments, the machine-readable storage medium 120 may include mass storage, removable storage, volatile read-write memory, read-only memory (ROM), and the like, or any combination thereof. Exemplary mass storage devices may include magnetic disks, optical disks, solid state disks, and the like. Exemplary removable memory may include flash drives, floppy disks, optical disks, memory cards, compact disks, magnetic tape, and the like. Exemplary volatile read-write memories can include Random Access Memory (RAM). Exemplary RAM may include healthy random access memory (DRAM), double data rate synchronous healthy random access memory (DDR SDRAM), static Random Access Memory (SRAM), thyristor random access memory (T-RAM), and zero capacitance random access memory (Z-RAM), among others. Exemplary read-only memories may include mask read-only memory (MROM), programmable read-only memory (PROM), erasable programmable read-only memory (perrom), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM), digital versatile disc read-only memory, and the like. In some embodiments, machine-readable storage medium 120 may be implemented on fixed point monitoring cloud platform 100. By way of example only, the fixed point monitoring cloud platform 100 may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-tiered cloud, and the like, or any combination thereof.

For a specific implementation process of the processor 110, reference may be made to the above-mentioned method embodiments executed by the fixed point monitoring cloud platform 100, and implementation principles and technical effects thereof are similar, and details of this embodiment are not described herein again.

In addition, the embodiment of the invention also provides a readable storage medium, wherein the readable storage medium is preset with computer-executable instructions, and when a processor executes the computer-executable instructions, the method for monitoring the fixed point of the construction site is realized.

It should be understood that the foregoing description is for purposes of illustration only and is not intended to limit the scope of the present disclosure. Many modifications and variations will be apparent to those of ordinary skill in the art in light of the description of the invention. However, such modifications and variations do not depart from the scope of the present invention.

While the basic concepts have been described above, it will be apparent to those of ordinary skill in the art in view of this disclosure that the above disclosure is intended to be exemplary only and is not intended to limit the invention. Various modifications, improvements and optimization of the invention will occur to those skilled in the art, although not explicitly described herein. Such modifications, improvements and optimized derivatives are proposed in the present invention and thus fall within the spirit and scope of the exemplary embodiments of this invention.

Also, the present invention has been described using specific terms to describe embodiments of the invention. For example, "one embodiment," "an embodiment," and/or "some embodiments" means that a particular feature, structure, or characteristic described in connection with at least one embodiment of the invention. Therefore, it is emphasized and should be appreciated that two or more references to "an embodiment" or "one embodiment" or "an alternative embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, certain features, structures, or characteristics of one or more embodiments of the invention may be combined as suitable.

Moreover, those skilled in the art will recognize that aspects of the present invention may be illustrated and described in terms of several patentable species or situations, including any new and useful process, machine, article, or material combination, or any new and useful modification thereof. Accordingly, aspects of the present invention may be embodied entirely in hardware, entirely in software (including firmware, resident software, micro-code, etc.) or in a combination of hardware and software. The above hardware or software may be referred to as a "unit", "module", or "system". Furthermore, aspects of the present disclosure may take the form of a computer program product embodied in one or more computer-readable media, with computer-readable program code embodied therein.

A computer readable signal medium may comprise a propagated data signal with computer program code embodied therein, for example, on a baseband or as part of a carrier wave. Such a propagated signal may take any of a variety of forms, including electro-magnetic, optical, and the like, or any suitable combination. A computer readable signal medium may be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code on a computer readable signal medium may be propagated over any suitable medium, including radio, electrical cable, fiber optic cable, RF, or the like, or any combination thereof.

Computer program code required for operation of various portions of the present invention may be written in any one or more of a variety of programming languages, including a subject oriented programming language such as Java, scala, smalltalk, eiffel, JADE, emerald, C + +, C #, VB.NET, python, and the like, a conventional programming language such as C, visual Basic, fortran 2003, perl, COBOL 2002, PHP, ABAP, a health programming language such as Python, ruby, and Groovy, or other programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any network format, such as a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet), or in a cloud computing environment, or as a service, such as a software as a service (SaaS).

Additionally, the order of processing elements and sequences, the use of numerical letters, or other designations herein, is not intended to limit the order of the processes and methods unless otherwise specified in the claims. While various presently contemplated useful embodiments of the invention have been discussed in the foregoing disclosure by way of examples, it is to be understood that such detail is solely for that purpose and that the appended claims are not intended to be limited to the disclosed embodiments, but, on the contrary, are intended to cover all modifications, equivalents, and combinations that are within the spirit and scope of the embodiments of the invention. For example, although the system components described above may be implemented by hardware devices, they may also be implemented by software-only solutions, such as installing the described system on an existing server or mobile device.

Similarly, it should be noted that in the foregoing description of embodiments of the invention, various features are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure aiding in the understanding of one or more of the embodiments. Similarly, it should be noted that in the preceding description of embodiments of the invention, various features are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure aiding in the understanding of one or more of the embodiments.

Claims

1. A building site fixed-point monitoring method is applied to a fixed-point monitoring cloud platform, the fixed-point monitoring cloud platform is in communication connection with a plurality of monitoring unmanned aerial vehicles, the monitoring unmanned aerial vehicles are used for fixed-point monitoring of fixed-point monitoring areas of a target building site, each fixed-point monitoring area corresponds to a different building logic function partition, and the method comprises the following steps:

carrying out violation behavior identification on each corresponding monitoring data stream based on the violation behavior identification network associated with the building logic function partition corresponding to each fixed point monitoring area, and obtaining violation behavior identification information of each corresponding monitoring data stream, wherein the method specifically comprises the steps of obtaining the violation behavior identification network associated with the building logic function partition corresponding to each fixed point monitoring area;

extracting target monitoring track characteristics of each corresponding monitoring data stream based on a violation behavior identification network associated with a building logic function partition corresponding to each fixed point monitoring area, and identifying violation behavior labels of the monitoring track characteristics to obtain violation behavior identification information of each corresponding monitoring data stream;

according to the violation behavior identification information of each corresponding monitoring data stream, early warning and reminding are carried out on the inspection on-duty terminal of each fixed-point monitoring area;

the method also comprises a step of training a violation behavior recognition network associated with the building logic function partition corresponding to each fixed point monitoring area, and the method comprises the following steps:

acquiring training monitoring track characteristics corresponding to training monitoring data streams of building logic function partitions corresponding to each fixed-point monitoring area, and inputting the training monitoring track characteristics to Q first-order AI training networks of a first AI training network to obtain Q illegal behavior identification information output by the Q first-order AI training networks; q first-order AI training networks are used for predicting Q kinds of violation behavior labels;

the first AI training network further comprises a W-order AI training network and a plurality of Q-1-order AI training networks, wherein the value of W is 2 to Q-2;

if the Q pieces of illegal behavior identification information correspond to R kinds of illegal behavior labels, and R is larger than 1 and smaller than Q, selecting a second AI training network from all R-order AI training networks of the first AI training network, wherein the second AI training network has a mapping relation with the R kinds of illegal behavior labels;

inputting the training monitoring track characteristics to a second AI training network to obtain violation behavior identification information output by the second AI training network, and determining a target violation behavior label corresponding to the training monitoring track characteristics based on the violation behavior identification information;

and determining a characteristic component sequence of the training monitoring track characteristic based on the target violation behavior label, performing network optimization on the first AI training network based on the training monitoring track characteristic and the characteristic component sequence, and determining a second AI training network based on the network-optimized AI training network, wherein the second AI training network is used for determining the violation behavior label corresponding to the monitoring data stream to be identified.

2. The construction site fixed-point monitoring method according to claim 1, wherein after inputting the training monitoring track characteristics to the Q first-order AI training networks of the first AI training network and obtaining the Q violation behavior recognition information output by the Q first-order AI training networks, the method further comprises:

and if the Q pieces of violation behavior identification information correspond to a violation behavior label, determining the violation behavior label as a target violation behavior label corresponding to the training monitoring track characteristic.

3. The construction site fixed-point monitoring method according to claim 2, wherein after inputting the training monitoring track characteristics to the Q first-order AI training networks of the first AI training network and obtaining the Q violation behavior recognition information output by the Q first-order AI training networks, the method further comprises:

if the Q illegal behavior identification information corresponds to Q illegal behavior labels, selecting the Q-1 order AI training network as a second AI training network; and inputting the training monitoring track characteristics to a second AI training network to obtain violation behavior identification information output by the second AI training network, and determining a target violation behavior label corresponding to the training monitoring track characteristics based on the violation behavior identification information.

4. The method according to claim 2, wherein the obtaining of the training and monitoring trajectory feature corresponding to the training and monitoring data stream of the building logic function partition corresponding to each fixed-point monitoring area comprises:

acquiring a training sample data set, wherein the training sample data set comprises monitoring track characteristics corresponding to training monitoring data streams of building logic function partitions corresponding to each fixed point monitoring area, and traversing one monitoring track characteristic from the training sample data set to serve as a training monitoring track characteristic;

the determining a second AI training network based on the network optimized AI training network includes:

determining whether the AI training network after network optimization meets training termination conditions;

if not, determining the network-optimized AI training network as a first AI training network, traversing another monitoring track feature from the training sample data set as a training monitoring track feature, and returning to execute the operation of inputting the training monitoring track feature to Q first-order AI training networks of the first AI training network;

and if so, determining a second AI training network based on the network optimized AI training network.

5. The method of claim 4, wherein determining the second AI training network based on the network optimized AI training network comprises:

determining the network-optimized AI training network as the second AI training network;

or selecting a Q-1 order AI training network from the network-optimized AI training networks, and determining a second AI training network based on the Q-1 order AI training network, wherein the second AI training network at least comprises the Q-1 order AI training network.

6. The construction site fixed point monitoring method according to claim 4, wherein the obtaining of the training monitoring trajectory feature corresponding to the training monitoring data stream of the building logic function partition corresponding to each fixed point monitoring area comprises:

acquiring a training monitoring data stream of a building logic function partition corresponding to each fixed point monitoring area, and extracting a plurality of monitoring characteristic configuration information aiming at the building logic function partition corresponding to each fixed point monitoring area to cluster a plurality of key track segments matched with current training monitoring data stream nodes at a first violation monitoring time period of the training monitoring data stream, wherein each key track segment cluster comprises a track characteristic field of the current training monitoring data stream node aiming at the monitoring characteristic configuration information;

determining a current key track segment characteristic set required by the current training monitoring data stream node from the building logic function partition according to the plurality of key track segment clusters;

when current key track segment characteristic sets respectively needed by a plurality of violation monitoring time sequence segments aiming at the current training monitoring data stream node are determined, generating a candidate key track segment clustering sequence of the current training monitoring data stream node corresponding to the building logic function partition by using key track segment contents respectively contained by the current key track segment characteristic sets, wherein the violation monitoring time sequence segments comprise the first violation monitoring time sequence segment and a violation monitoring time sequence segment behind the first violation monitoring time sequence segment;

when a plurality of candidate key track segment clustering sequences generated by a reference violation monitoring time sequence segment corresponding to the building logic function partition are obtained, screening the candidate key track segment clustering sequences to obtain a target key track segment clustering sequence;

extracting the characteristics of the target key track segment clustering sequence to obtain the characteristics of the training monitoring track;

wherein the determining a current critical trajectory segment feature set required by the current training monitoring data stream node from the building logical function partition according to the plurality of critical trajectory segment clusters includes:

and determining the matched key track segment characteristics corresponding to the key track segment matching template corresponding to the maximum candidate matching degree in all the candidate matching degrees as the current key track segment characteristic set of the current training monitoring data stream node in the first violation monitoring time period.

7. The construction site fixed-point monitoring method according to claim 5, wherein after the AI training network based on network optimization determines the second AI training network, the step of extracting the target monitoring track characteristics of each corresponding monitoring data stream based on the violation behavior recognition network associated with the building logic function partition corresponding to each fixed-point monitoring area, recognizing the violation behavior labels for the monitoring track characteristics, and obtaining the violation behavior recognition information of each corresponding monitoring data stream comprises:

if the second AI training network comprises Q first-order AI training networks, a plurality of W-order AI training networks and a plurality of Q-1-order AI training networks, after the target monitoring track characteristics corresponding to each corresponding monitoring data stream to be identified are obtained, inputting the target monitoring track characteristics to the Q first-order AI training networks of the second AI training network to obtain Q illegal behavior identification information output by the Q first-order AI training networks;

if the Q pieces of violation behavior identification information correspond to a violation behavior label, determining the violation behavior label as a target violation behavior label corresponding to the target monitoring track characteristic;

if the Q pieces of illegal behavior identification information correspond to R types of illegal behavior labels, selecting a second AI training network from all R-order AI training networks of the second AI training network, inputting the target monitoring track characteristics to the second AI training network to obtain illegal behavior identification information output by the second AI training network, and determining the target illegal behavior label corresponding to the target monitoring track characteristics based on the illegal behavior identification information;

if the Q pieces of illegal behavior identification information correspond to Q kinds of illegal behavior labels, selecting the Q-1 order AI training network as a second AI training network, inputting the target monitoring track characteristics to the second AI training network to obtain illegal behavior identification information output by the second AI training network, and determining the target illegal behavior label corresponding to the target monitoring track characteristics based on the illegal behavior identification information;

or if the second AI training network comprises a plurality of Q-1 order AI training networks, after obtaining the target monitoring track characteristics corresponding to the monitoring data stream to be identified, inputting the target monitoring track characteristics to the Q-1 order AI training network of the second AI training network to obtain the violation behavior identification information output by the Q-1 order AI training network;

and determining a target violation behavior label corresponding to the target monitoring track characteristic based on the violation behavior identification information.

8. A building site fixed-point monitoring system is characterized by comprising a fixed-point monitoring cloud platform and a plurality of monitoring unmanned aerial vehicles in communication connection with the fixed-point monitoring cloud platform, wherein the monitoring unmanned aerial vehicles are used for fixed-point monitoring of fixed-point monitoring areas of a target building site, and each fixed-point monitoring area corresponds to a different building logic function partition;

the fixed point monitoring cloud platform is used for:

carrying out violation behavior identification on each corresponding monitoring data stream based on a violation behavior identification network associated with the building logic function partition corresponding to each fixed point monitoring area to obtain violation behavior identification information of each corresponding monitoring data stream; acquiring a violation behavior identification network associated with a building logic function partition corresponding to each fixed point monitoring area;

the fixed point monitoring cloud platform is further used for:

training the violation behavior recognition network associated with the building logic function partition corresponding to each fixed point monitoring area, comprising the following steps: