An AI-based intelligent supervision and inspection system

Abstract

This invention belongs to the field of engineering supervision and inspection technology, and discloses an AI-based intelligent supervision and inspection system. The system includes an inspection item configuration unit, an image acquisition unit, a reference positioning unit, an identification unit, a judgment unit, a supplementary image generation unit, a re-inspection positioning unit, and a re-inspection verification unit. The system calls the corresponding judgment rules based on the inspection item identifier, acquires inspection images of the supervision area, identifies benchmark reference objects, target objects, and necessary observation points, establishes relative position parameters, and outputs qualified judgment results, unqualified judgment results, or supplementary image identifiers accordingly. When an item is unqualified, re-inspection positioning information is generated, and the re-inspection area is determined in the re-inspection image to complete the re-inspection verification. This system is applicable to the judgment and re-inspection of inspection images under different equipment and angles.

Description

Technical Field

[0001] This invention belongs to the field of engineering supervision and inspection technology, specifically an AI-based intelligent supervision and inspection system. Background Technology

[0002] In the process of construction project supervision, it is usually necessary to continuously inspect the construction site, focusing on items such as opening protection, power distribution facilities, edge protection, and equipment installation status, to determine whether the site layout meets the corresponding requirements. With the increase in construction areas, the dispersion of work surfaces, and the increase in inspection frequency, on-site supervision has gradually shifted to acquiring inspection images using fixed camera equipment, handheld acquisition terminals, mobile inspection equipment, or unmanned inspection equipment, and then combining them with the back-end system to complete recording, analysis, and review. Inspection processing based on image information has become a common technical application method.

[0003] In existing technologies, inspection systems typically acquire on-site images based on preset inspection areas, use image recognition models to identify target objects, and then output inspection results in conjunction with preset rules. Some solutions may simultaneously identify multiple objects or call multiple images of the same area for auxiliary judgment. After an anomaly is detected, a rectification record will be generated, and on-site images will be reacquired during subsequent reviews to determine whether the status has changed after rectification. Thus, building an inspection process around image acquisition, object recognition, rule judgment, and review verification has become a common processing mode for existing smart supervision systems.

[0004] However, inspection images acquired at different times and with different equipment at the construction site often exhibit variations in shooting direction, pitch angle, obstruction, and framing range. This can lead to shifts in the positional representation of the same target object and related observation points in different images. Particularly in inspection items that rely on relative positional relationships for judgment, such as opening protection, door closure, and component installation, even if the existing system can identify the target object or local observation points, it may still be difficult to reliably determine whether the target object and the corresponding observation point meet the required positional relationship due to incomplete exposure of necessary observation points and a lack of unified positional references between different images. Furthermore, if anomalies have been recorded in the previous inspection, subsequent review images may lack consistent positional references and regional correspondences with the previous inspection, easily leading to inconsistencies between the reviewed and initially determined areas. This affects the correspondence between the results of the two inspections and the reliability of the review judgment. Summary of the Invention

[0005] The purpose of this invention is to provide an AI-based intelligent supervision and inspection system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an AI-based intelligent supervision and inspection system for judging and subsequently re-inspecting and locating inspection items within a predetermined supervision area based on inspection images.

[0007] The system includes an inspection item configuration unit, an image acquisition unit, a reference positioning unit, an identification unit, a judgment unit, and a re-inspection positioning unit. The inspection item configuration unit receives the identifier of the inspection item to be executed, retrieves the judgment rule corresponding to that inspection item, and determines the supervision area identifier corresponding to that inspection item, ensuring that subsequent image acquisition and judgment processes are confined to the corresponding supervision area. The image acquisition unit acquires inspection images of the corresponding supervision area based on the supervision area identifier and records acquisition information such as acquisition time, acquisition location, and shooting parameters, providing an image basis and acquisition reference for subsequent identification and judgment. The reference positioning unit identifies benchmark reference objects located within the supervision area from the inspection images and establishes reference relationships within the supervision area, providing a reference basis for determining the location of target objects and necessary observation points. The identification unit retrieves the image identifier corresponding to the inspection item identifier. The identification model identifies target objects and necessary observation points in the inspection image, and determines the position parameters of the target objects and necessary observation points relative to the reference reference object based on the reference relationship, and outputs the target object identification result and the necessary observation point identification result. The judgment unit is used to combine the target object identification result, the necessary observation point identification result, the position parameters and the judgment rules to output the qualified judgment result, unqualified judgment result or the pending re-sampling mark for the inspection item corresponding to the current inspection image, so as to distinguish whether the current inspection image has met the conditions for completing the judgment of the inspection item. The re-inspection positioning unit is used to record the position parameters of the target object that generated the unqualified judgment result relative to the reference reference object and the corresponding shooting parameters when the judgment unit outputs the unqualified judgment result, and generate re-inspection positioning information so that the corresponding area can be re-positioned and verified based on the position and shooting conditions corresponding to the previous judgment in the subsequent re-inspection process.

[0008] In the aforementioned AI-based intelligent supervision and inspection system, the image acquisition unit records acquisition information including acquisition time, acquisition device identification, acquisition location, and shooting parameters, wherein the shooting parameters include one or more of shooting direction, pitch angle, and focal length. The image acquisition unit acquires inspection images through fixed camera equipment, handheld acquisition terminals, mobile inspection equipment, or unmanned inspection equipment, so that inspection images acquired in different supervision areas can be used as input data for subsequent identification and judgment.

[0009] To ensure consistent positional references across different inspection images, the system uses objects within the same supervision area that maintain stable positions and can be repeatedly identified in different inspection images as benchmark references. These benchmark references include one or more of the following: structural edge lines, column outlines, beam-slab intersection lines, opening boundaries, equipment installation bases, or positioning markers.

[0010] After establishing a location reference base within the supervision area based on the aforementioned benchmark, the distance parameters, orientation parameters, and adjacency parameters of the target object and necessary observation points relative to the benchmark can be determined accordingly, providing a unified location basis for the judgment of the same inspection item under different inspection images.

[0011] Furthermore, the inspection item configuration unit pre-stores the correspondence between multiple inspection item identifiers and multiple judgment rules; each inspection item identifier corresponds to a set of target objects, a set of necessary observation points, a set of target object existence requirements, a set of preset positional relationships, and a set of minimum visibility requirements.

[0012] The existence requirements of the target object include one or more of the following: should exist, should not exist, and the quantity meets the preset quantity requirements. After receiving the inspection item identifier to be executed, the inspection item configuration unit provides the corresponding target object and necessary observation points to the identification unit, and provides the corresponding target object existence requirements, preset positional relationships, and minimum visibility requirements to the judgment unit.

[0013] With the above settings, different inspection items correspond to different identification objects and judgment criteria, so that each inspection item can be processed according to its own supervision requirements.

[0014] During the recognition process, the recognition unit uses one or more of the following models: target detection model, semantic segmentation model, key point recognition model, and image matching model to jointly recognize the target object and necessary observation points. After recognition is completed, the recognition unit combines the position parameters of the target object and the necessary observation points relative to the same reference object to establish the association between the target object and the necessary observation points, and outputs the recognition results of the target object and the recognition results of the necessary observation points.

[0015] The minimum visibility requirement is set as a visibility requirement that can be used to determine the preset positional relationship between the target object and the necessary observation point; the determination unit outputs the corresponding determination result based on the target object identification result, the necessary observation point identification result, the correlation relationship and the determination rule.

[0016] Specifically, when both the target object and the necessary observation point are identified, the necessary observation point meets the minimum visibility requirement, the existence of the target object meets the target object existence requirement, and the target object and the necessary observation point satisfy the preset positional relationship, a qualified judgment result is output; when the necessary observation point meets the minimum visibility requirement, but the existence of the target object does not meet the target object existence requirement, or the target object and the necessary observation point do not satisfy the preset positional relationship, a unqualified judgment result is output; when the necessary observation point does not meet the minimum visibility requirement, or the association relationship between the target object and the necessary observation point is not established, a supplementary sampling indicator is output.

[0017] Using the above-mentioned methods, the missing target and the location not matching the evidence can be distinguished, thus providing a clear basis for subsequent supplementary sampling or re-examination.

[0018] In some implementations, the determination of the same inspection item can be completed by one inspection image or by multiple inspection images obtained in the same inspection for the same supervision area. When multiple inspection images are used to complete the determination, the reference positioning unit establishes the correspondence between the images based on the reference objects in each inspection image. The determination unit then associates the target objects and necessary observation points in each inspection image based on the correspondence between the images, and outputs the qualified determination result, the unqualified determination result, or the mark to be supplemented.

[0019] When the above method is adopted, if a single inspection image has a limited viewing angle or partial obstruction, other inspection images in the same inspection can be used to supplement the necessary observation points or the observation content of the target object, so as to ensure the integrity of the judgment basis.

[0020] When the judgment unit outputs the identifier to be supplemented, the system further generates supplementary sampling information; specifically, the system determines the area to be supplemented, the shooting direction to be supplemented, and the object to be confirmed based on the necessary observation points that have not met the minimum visibility requirements, or the positional relationship of the target object that has not been established with the reference object relative to the reference object.

[0021] The objects to be confirmed include one or both of the necessary observation points and target objects. Through the above settings, supplementary data collection is carried out around the missing observation content, thereby avoiding repeated data collection on the entire supervision area.

[0022] When the judgment unit outputs a non-conformance judgment result, the re-inspection positioning unit records the position parameters and shooting parameters of the target object that generated the non-conformance judgment result relative to the reference object, and generates re-inspection positioning information. After acquiring the re-inspection image, the re-inspection verification unit first calls the corresponding judgment rule according to the inspection item identifier in the re-inspection positioning information, then identifies the same reference object as before rectification in the re-inspection image, and determines the re-inspection area corresponding to the non-conformance judgment result based on the position parameters and shooting parameters of the target object relative to the reference object in the re-inspection positioning information.

[0023] Subsequently, the re-inspection and verification unit identifies the target objects and necessary observation points within the re-inspection area, and outputs a re-inspection pass result or a re-inspection fail result based on the identification results and judgment rules. Through the above settings, the initial judgment and subsequent re-inspection can be carried out around the same target area, so that the two verifications have consistent location basis and judgment basis.

[0024] The beneficial effects of this invention are as follows: 1. This invention, through the cooperation of an inspection item configuration unit, a reference positioning unit, and an identification unit, pre-assigns the target object, necessary observation points, and their positional relationship requirements corresponding to the inspection item to the specific supervision area, and introduces a benchmark reference object as a unified positional reference basis in the inspection image. In this way, between inspection images acquired by different devices, from different angles, and at different times, the target object and necessary observation points can all be expressed in position and their relationships judged around the same benchmark reference object, no longer relying solely on the local appearance in a single image to draw conclusions. Therefore, for inspection items such as opening protection, door closure, and component installation that rely on relative positional relationships for judgment, a relatively stable judgment basis can be maintained even when the image framing range or shooting direction changes, thereby reducing misjudgments and omissions caused by positional representation shifts.

[0025] 2. This invention does not directly output a conclusion after identifying a single object. Instead, it incorporates the minimum visibility requirements of necessary observation points, the existence requirements of the target object, and the preset positional relationship between the target object and the necessary observation points into the judgment process. Based on this, it sets up a supplementary sampling marker and a supplementary sampling generation unit. Thus, when a partial object is identified in the inspection image, but the necessary observation points are not fully revealed, or a stable relationship has not yet been established between the target object and the necessary observation points, the system will not confuse the case of insufficient evidence with the case of non-compliance. Instead, it will further generate a supplementary sampling area, a supplementary sampling shooting direction, and an object to be confirmed corresponding to the missing observation content. As a result, supplementary sampling can be carried out around the missing evidence, so that the subsequent judgment is based on a more complete set of observation content, thereby improving the utilization of inspection images under complex field conditions.

[0026] 3. This invention records the position and shooting parameters of the target object relative to the reference object when the initial non-conformity judgment result is output. During subsequent re-inspections, the re-inspection verification unit determines the re-inspection area based on these parameters and then completes the re-inspection verification by combining the judgment rules of the same inspection item. In this way, the previous inspection and subsequent re-inspections revolve around the same target area, the same reference base, and the same judgment basis, making it less likely for the re-inspection area to deviate from the initial judgment area. Especially under conditions where there are diverse image sources at the construction site and changes in shooting angles before and after rectification, the correspondence between the two verification results can still be maintained, making the re-inspection judgment closer to the actual rectification object, thereby improving the pertinence and consistency of the abnormality re-inspection process. Attached Figure Description

[0027] Figure 1 This is a flowchart of the AI-based intelligent supervision and inspection system of the present invention; Figure 2 This is a flowchart for determining the inspection results of this invention; Figure 3 This is a flowchart of the re-inspection and verification process for non-conforming items in this invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] like Figures 1 to 3 As shown, this embodiment of the invention provides an AI-based intelligent supervision and inspection system, including an inspection item configuration unit, an image acquisition unit, a reference positioning unit, an identification unit, a judgment unit, and a re-inspection positioning unit.

[0030] The inspection item configuration unit is used to receive the inspection item identifier to be executed and retrieve the corresponding judgment rules and supervision area identifier according to the inspection item identifier; the image acquisition unit is used to acquire the inspection image of the corresponding supervision area according to the supervision area identifier and record the acquisition time, acquisition location, acquisition device identifier and shooting parameters corresponding to the inspection image.

[0031] The reference positioning unit is used to identify the benchmark reference objects located within the supervision area from the inspection image and establish the position reference basis of the supervision area; the identification unit is used to call the image recognition model corresponding to the inspection item identifier, identify the target object and necessary observation point in the inspection image, and determine the position parameters of the target object and necessary observation point relative to the benchmark reference object in combination with the position reference basis, and output the target object identification result and the necessary observation point identification result.

[0032] The judgment unit is used to judge the inspection items corresponding to the current inspection image based on the target object recognition result, the necessary observation point recognition result, the position parameters and the judgment rules, and output the qualified judgment result, the unqualified judgment result or the pending re-collection mark; the re-inspection positioning unit is used to record the position parameters and shooting parameters of the target object that generated the unqualified judgment result relative to the reference object when the judgment unit outputs the unqualified judgment result, and generate re-inspection positioning information for subsequent re-inspection to locate the corresponding area.

[0033] In this embodiment, an AI-based intelligent supervision and inspection system includes an inspection item configuration unit, an image acquisition unit, a reference positioning unit, an identification unit, a judgment unit, a supplementary image generation unit, a re-inspection positioning unit, and a re-inspection verification unit. The inspection item configuration unit, reference positioning unit, identification unit, judgment unit, supplementary image generation unit, re-inspection positioning unit, and re-inspection verification unit are deployed on an inspection processing server. The image acquisition unit consists of on-site acquisition devices, including one or more of fixed camera equipment, handheld acquisition terminals, mobile inspection equipment, and unmanned inspection equipment. When inspection images from different sources are accessed by the inspection processing server, they are accompanied by corresponding acquisition information, which serves as the basis for subsequent identification, judgment, and re-inspection positioning.

[0034] In this embodiment, the inspection item identifier is used to uniquely indicate an inspection item to be executed, and the supervision area identifier is used to uniquely indicate a supervision area in the construction site; each inspection item identifier is pre-established with one or more supervision area identifiers, so that the system can determine the supervision area corresponding to this inspection after receiving the inspection item identifier.

[0035] The inspection item configuration unit pre-stores the correspondence between multiple inspection item identifiers and multiple judgment rules; each judgment rule is established around a single inspection item, including the target object, necessary observation points, the existence requirements of the target object, the preset positional relationship between the target object and the necessary observation points, and the minimum visibility requirements of the necessary observation points.

[0036] The target object existence requirement specifies whether the target object should exist, should not exist, or its quantity meets the preset quantity requirement under the current inspection item; the preset positional relationship specifies the positional constraints that the target object should satisfy relative to the necessary observation point; the minimum visibility requirement specifies the visibility level at which the system can determine whether the target object and the necessary observation point satisfy the preset positional relationship.

[0037] The aforementioned judgment rules are configured before system deployment. During configuration, based on the inspection forms, construction site templates, and existing inspection images in actual supervision work, the target objects and necessary observation points corresponding to each inspection item are determined. Then, the target objects, necessary observation points, and benchmark references in the template images and on-site inspection images are labeled to form training samples for the image recognition model and the basis for configuring the judgment rules. In this way, both the image recognition model and the judgment rules are established around specific inspection items.

[0038] In this embodiment, the acquisition information recorded by the image acquisition unit includes acquisition time, acquisition device identification, acquisition location, and shooting parameters; the shooting parameters include one or more of shooting direction, pitch angle, and focal length; fixed camera equipment is used to acquire continuous inspection images of the supervised area, handheld acquisition terminal is used to acquire close-range inspection images, mobile inspection equipment is used to acquire inspection images during movement along a predetermined inspection route, and unmanned inspection equipment is used to acquire inspection images of high places, edges, or areas inaccessible to humans; after the inspection images acquired by different devices enter the inspection processing server, the acquisition time, acquisition location, and shooting parameters are stored in a unified field format, so that subsequent reference positioning, identification, and re-inspection positioning can all call up the same type of acquisition information.

[0039] In this embodiment, the reference positioning unit identifies objects within the supervision area that maintain a stable position and can be repeatedly identified in different inspection images as reference objects. The reference objects include one or more of the following: structural edge lines, column outlines, beam-slab intersection lines, opening boundaries, equipment installation bases, and positioning markers. After identifying the reference objects, the reference positioning unit extracts the outline lines, intersection points, boundary points, or center positions of the reference objects to establish the positional reference basis for the supervision area.

[0040] The position reference base is used for subsequent calculations of the position parameters of the target object and necessary observation points relative to the reference object.

[0041] In this embodiment, the identification unit calls the image recognition model corresponding to the inspection item identifier to jointly identify the target object and necessary observation points in the inspection image; the image recognition model includes one or more of the following: target detection model, semantic segmentation model, key point recognition model and image matching model.

[0042] Specifically, for target objects with complete outer contours, a target detection model is used to identify the outer region of the target object; for target objects with clear boundaries and requiring differentiation of coverage areas, a semantic segmentation model is used to extract their contour regions; for necessary observation points that need to be extracted, such as endpoints, inflection points, or connection points, a key point recognition model is used to extract key positions; and for objects with relatively fixed shapes that need to be locally compared with template images, an image matching model is used for matching and recognition.

[0043] After identification is completed, the identification unit determines the distance parameters, orientation parameters, and adjacency parameters of the target object and the necessary observation point relative to the same reference object. Furthermore, when the position parameters of the target object and the necessary observation point relative to the same reference object meet the association conditions preset for the corresponding inspection item, the identification unit establishes the association relationship between the target object and the necessary observation point, and outputs the identification results of the target object and the necessary observation point.

[0044] In this embodiment, the minimum visibility requirement does not merely require the target object or necessary observation point to appear in the inspection image, but rather requires that the necessary observation point reach a level of visibility sufficient to determine a preset positional relationship. For linear necessary observation points, the minimum visibility requirement is that its continuous edges reach a preset visible length and its extension direction can be determined; for point-like necessary observation points, the minimum visibility requirement is that its key location points can be identified and its positional relationship with a reference object can be established; for area-like necessary observation points, the minimum visibility requirement is that its boundary region reaches a preset visible integrity and its relative coverage relationship with the target object can be determined. If a necessary observation point is roughly identified but does not meet the above requirements, it is not considered to satisfy the minimum visibility requirement.

[0045] In this embodiment, the determination unit outputs a qualified determination result, an unqualified determination result, or a pending sampling mark for the inspection item corresponding to the current inspection image based on the target object identification result, the necessary observation point identification result, the position parameters, and the determination rules.

[0046] The specific judgment process is as follows: When both the target object and the necessary observation point are identified, the necessary observation point meets the minimum visibility requirement, the existence of the target object meets the target object existence requirement of the corresponding inspection item, and the target object and the necessary observation point satisfy the preset positional relationship of the corresponding inspection item, the judgment unit outputs a qualified judgment result; when the necessary observation point meets the minimum visibility requirement, but the existence of the target object does not meet the target object existence requirement, or the target object and the necessary observation point do not satisfy the preset positional relationship, the judgment unit outputs a unqualified judgment result; when the necessary observation point does not meet the minimum visibility requirement, or although the target object and the necessary observation point are identified, but the association between the two is not established, the judgment unit outputs a supplementary sampling indicator.

[0047] In this embodiment, the determination of the same inspection item is completed by one inspection image, or by multiple inspection images obtained in the same inspection for the same supervision area.

[0048] The same inspection refers to the image acquisition process completed within the preset time window of the same inspection task for the same inspection item and the same supervision area. If an inspection image has completely contained the target object and the necessary observation points and meets the minimum visibility requirements for judgment, the judgment can be completed directly from that single inspection image.

[0049] If a single inspection image contains partial obstruction, skewed viewpoint, or the target object and necessary observation points are scattered in different images, then multiple inspection images obtained in the same inspection for the same supervised area will be used to make a joint judgment.

[0050] At this point, the reference positioning unit establishes the correspondence between images based on the reference objects in each inspection image. The judgment unit then associates the target objects and necessary observation points in each inspection image based on the correspondence between images, and then outputs the qualified judgment result, the unqualified judgment result, or the mark to be supplemented.

[0051] In this embodiment, when the determination unit outputs a marker to be supplemented, the supplementary acquisition generation unit generates supplementary acquisition information based on the positional relationship between the necessary observation points that have not met the minimum visibility requirement or the target objects that have not established a relationship and the reference reference. The supplementary acquisition information includes the area to be supplemented, the shooting direction to be supplemented, and the object to be confirmed. The object to be confirmed includes one or both of the necessary observation points and the target object. The supplementary acquisition generation unit first determines the area to be supplemented based on the positional parameters of the object to be confirmed relative to the reference reference, and then determines the shooting direction to be supplemented based on the missing or occluded direction of the object to be confirmed in the inspection image, so that the supplementary acquisition revolves around the missing observation content.

[0052] In this embodiment, when the judgment unit outputs a non-compliance judgment result, the re-inspection positioning unit records the position parameters and shooting parameters of the target object that generated the non-compliance judgment result relative to the reference object, and generates re-inspection positioning information; the re-inspection positioning information includes at least the inspection item identifier, the supervision area identifier, the distance parameters, orientation parameters, and adjacency parameters of the target object relative to the reference object, as well as one or more shooting parameters among the shooting direction, pitch angle and focal length.

[0053] In this embodiment, after acquiring the re-inspection image, the re-inspection verification unit first calls the corresponding judgment rule based on the inspection item identifier in the re-inspection positioning information, and then identifies the same reference object as before rectification in the re-inspection image. After identifying the reference object, the re-inspection verification unit uses the reference object as the position reference basis in the re-inspection image, and then determines the re-inspection area based on the position parameters and shooting parameters of the target object relative to the reference object in the re-inspection positioning information. It then identifies the target object and necessary observation points in the re-inspection area, and outputs the re-inspection qualified result or the re-inspection failed result according to the identification result and the judgment rule.

[0054] If the existence of the target object in the re-inspection image meets the corresponding requirements, and the target object and the necessary observation point satisfy the preset positional relationship, then the re-inspection result is output as qualified; if the existence of the target object still does not meet the requirements, or the target object and the necessary observation point still do not satisfy the preset positional relationship, then the re-inspection result is output as unqualified.

[0055] In this embodiment, the inspection item configuration, image acquisition, benchmark reference positioning, target object identification, judgment result output, supplementary data acquisition and re-inspection verification form a sequential processing chain. This processing chain revolves around the target object, necessary observation points, positional relationships and re-inspection positioning information corresponding to the inspection item, so that the same inspection item has consistent judgment and positioning basis in the initial judgment and subsequent re-inspection processes.

[0056] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0057] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI-based intelligent supervision and inspection system, characterized in that: It includes an inspection item configuration unit, an image acquisition unit, a reference positioning unit, an identification unit, a judgment unit, and a re-inspection positioning unit; The inspection item configuration unit is used to receive the inspection item identifier to be executed, call the judgment rule corresponding to the inspection item identifier, and output the supervision area identifier corresponding to the inspection item identifier. The image acquisition unit is used to acquire inspection images of the corresponding supervision area according to the supervision area identifier, and record the acquisition information corresponding to the inspection image; The reference positioning unit is used to identify the reference objects located within the supervision area from the inspection image, and to establish the position parameters of the target object and necessary observation points relative to the reference objects; The recognition unit is used to call the image recognition model corresponding to the inspection item identifier, identify the target object and necessary observation point in the inspection image, and output the target object recognition result and the necessary observation point recognition result; The determination unit is used to output a qualified determination result, an unqualified determination result, or a pending re-collection mark for the inspection item corresponding to the current inspection image based on the target object recognition result, the necessary observation point recognition result, the position parameter, and the determination rule. The re-inspection positioning unit is used to record the position parameters and shooting parameters of the corresponding target object relative to the reference object when the judgment unit outputs the non-conformity judgment result, so as to generate re-inspection positioning information.

2. The AI-based intelligent supervision and inspection system according to claim 1, characterized in that: The collected information includes one or more of the following: collection time, collection device identifier, collection location, and shooting parameters; The shooting parameters include one or more of the following: shooting direction, pitch angle, and focal length; The image acquisition unit acquires the inspection images through fixed camera equipment, handheld acquisition terminal, mobile inspection equipment, or unmanned inspection equipment.

3. The AI-based intelligent supervision and inspection system according to claim 2, characterized in that: The reference object is an object whose position remains stable within the same supervision area and can be repeatedly identified in different inspection images; The reference objects include one or more of the following: structural edge lines, column outlines, beam-slab intersection lines, opening boundaries, equipment installation bases, and positioning marks. The position parameters include one or more of the following: distance parameters, orientation parameters, and adjacency parameters of the target object and necessary observation points relative to the reference reference.

4. The AI-based intelligent supervision and inspection system according to claim 3, characterized in that: The inspection item configuration unit pre-stores the correspondence between multiple inspection item identifiers and multiple judgment rules; Each inspection item identifier corresponds to a set of target objects, a set of necessary observation points, a set of target object existence requirements, a set of preset positional relationships, and a set of minimum visibility requirements; The existence requirement of the target object includes one or more of the following: should exist, should not exist, and the quantity meets the preset quantity requirement; The inspection item configuration unit sends the corresponding target object and necessary observation point to the identification unit according to the received inspection item identifier, and sends the corresponding target object existence requirements, preset positional relationships and minimum visibility requirements to the determination unit.

5. The AI-based intelligent supervision and inspection system according to claim 4, characterized in that: The image recognition model includes one or more of the following: object detection model, semantic segmentation model, key point recognition model, and image matching model; The identification unit performs joint identification of the target object and the necessary observation point, and establishes the association between the target object and the necessary observation point based on the position parameters of the target object and the necessary observation point relative to the same reference object, and then outputs the identification result of the target object and the identification result of the necessary observation point.

6. The AI-based intelligent supervision and inspection system according to claim 5, characterized in that: The minimum visibility requirement is the visibility requirement that enables the determination of the preset positional relationship between the target object and the necessary observation point.

7. The AI-based intelligent supervision and inspection system according to claim 6, characterized in that: The determination unit outputs the determination result according to the following rules: When both the target object and the necessary observation point are identified, the necessary observation point meets the minimum visibility requirement, the existence of the target object meets the target object existence requirement, and the target object and the necessary observation point satisfy the preset positional relationship, the qualified judgment result is output. When the minimum visibility requirement is met at the necessary observation point, but the existence of the target object does not meet the target object existence requirement, or the target object and the necessary observation point do not satisfy the preset positional relationship, the unqualified judgment result is output. If the minimum visibility requirement is not met at the necessary observation point, or if the association between the target object and the necessary observation point is not established, the identifier to be supplemented for sampling is output.

8. The AI-based intelligent supervision and inspection system according to claim 7, characterized in that: The determination of the same inspection item mark is completed by one inspection image, or by multiple inspection images obtained in the same inspection for the same supervision area. When multiple inspection images are used to make a judgment, the reference positioning unit establishes the correspondence between the images based on the reference objects in each inspection image; Based on the correspondence between the images, the determination unit associates the target objects and necessary observation points in each inspection image, and then outputs the qualified determination result, the unqualified determination result, or the identifier to be supplemented.

9. The AI-based intelligent supervision and inspection system according to claim 8, characterized in that: It also includes a supplementary sampling generation unit; The supplementary sampling generation unit is used to generate supplementary sampling information relative to the reference reference object based on the positional relationship of the target object that has not met the minimum visibility requirement or has not established the association relationship when the determination unit outputs the supplementary sampling identifier. The supplementary sampling information includes one or more of the following: the area to be supplemented, the shooting direction to be supplemented, and the object to be confirmed; the object to be confirmed includes one or two of the following: necessary observation points and target objects.

10. The AI-based intelligent supervision and inspection system according to claim 9, characterized in that: It also includes a re-inspection and verification unit; The re-inspection and verification unit is used to call the corresponding judgment rule according to the inspection item identifier in the re-inspection positioning information after acquiring the re-inspection image, and to identify the same benchmark reference object as before rectification in the re-inspection image; The re-inspection and verification unit determines the re-inspection area corresponding to the non-compliance judgment result based on the position parameters and shooting parameters of the target object relative to the reference object in the re-inspection positioning information. The re-inspection and verification unit identifies the target objects and necessary observation points within the re-inspection area, and outputs a re-inspection pass result or a re-inspection fail result based on the identification results and the judgment rules.

Images