Hazard report generation methods, devices, equipment, and media

Abstract

This application relates to a method, apparatus, device, and medium for generating hazard reports. The method includes: acquiring an original work image sent by target glasses; sending the original work image to a preset cloud server, requesting the cloud server to call a preset hazard identification model to perform hazard detection on the original work image, and obtaining a hazard identification result; upon receiving the hazard identification result returned by the cloud server, embedding work metadata into the original work image based on a preset watermark template to generate a verifiable graphic result; and generating a target hazard report based on the hazard identification result and the verifiable graphic result. This application effectively integrates the hazard identification result with the original work image embedded with an anti-tampering watermark to generate a hazard report that includes intuitive image evidence and verifiable authenticity, significantly improving the credibility of the report and management efficiency.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, device, and medium for generating hazard reports. Background Technology

[0002] In industrial production, construction, and power maintenance, the timely detection and accurate recording of on-site safety hazards are crucial for ensuring production safety. With the development of smart wearable devices and artificial intelligence technology, using wearable devices to collect images of the work site and then identifying hazards through cloud-based AI models has become an important technical means to improve the efficiency of hazard identification.

[0003] In existing technologies, workers typically use fixed cameras or handheld mobile devices to capture images of the work site. These images are then uploaded to a cloud server, where an AI model analyzes them to identify potential safety hazards. After hazard identification, the cloud server returns the results to the workers or managers, who then manually compile a hazard report based on the findings. While this approach automates hazard identification, it still has significant shortcomings in the hazard report generation process.

[0004] Specifically, existing hazard reports typically only contain textual descriptions of hazard identification results, failing to effectively integrate these results with original work images. This results in a lack of intuitive visual evidence, making it difficult for managers to quickly pinpoint the exact location and condition of hazards at the work site. More critically, existing hazard reports lack effective anti-tampering mechanisms. Original work images may be maliciously modified or replaced during transmission, storage, and report generation, and current solutions cannot verify the authenticity of the images, making it difficult to guarantee the credibility of hazard reports. Summary of the Invention

[0005] The primary objective of this application is to address at least one of the aforementioned problems by providing a method, apparatus, equipment, or medium for generating hazard reports.

[0006] To achieve the various objectives of this application, the following technical solution is adopted: A method for generating a hazard report, provided for one of the purposes of this application, includes the following steps: Acquire the raw image of the task sent by the target glasses; The original work image is sent to a preset cloud server, which requests the cloud server to call a preset hazard identification model to perform hazard detection on the original work image and obtain the hazard identification result. When the hazard identification result returned by the cloud server is received, the operation metadata is embedded into the original operation image based on the preset watermark template to generate a verifiable graphic result; Based on the hazard identification results and the verifiable graphic results, a target hazard report is generated.

[0007] A hazard report generation apparatus, proposed to meet one of the purposes of this application, is a hazard report generation method, comprising: The image acquisition module is configured to acquire the raw job image sent by the target glasses; The hazard identification module is configured to send the original work image to a preset cloud server, request the cloud server to call a preset hazard identification model to perform hazard detection on the original work image, and obtain the hazard identification result; The watermark embedding module is configured to embed the operation metadata into the original operation image based on a preset watermark template when the hazard identification result returned by the cloud server is received, thereby generating a verifiable image and text result. The report generation module is configured to generate a target hazard report based on the hazard identification results and the verifiable graphic results.

[0008] In another aspect, a computer device provided for one of the purposes of this application includes a central processing unit and a memory, the central processing unit being used to invoke and run a computer program stored in the memory to perform the steps of the hazard report generation method described in this application.

[0009] On another aspect, a computer-readable storage medium is provided to suit another purpose of this application, which stores in the form of computer-readable instructions a computer program implemented according to the described hazard report generation method, which, when called by a computer, executes the steps included in the corresponding method.

[0010] Compared with existing technologies, the advantages of this application are as follows: First, this application identifies potential hazards by sending the original work images to a cloud server. Upon receiving the identification results, it embeds work metadata into the original work images based on a preset watermark template to generate verifiable text and image results. Finally, it generates a target hazard report based on the hazard identification results and the verifiable text and image results, achieving deep integration of hazard identification results and original work images. This application integrates the hazard information identified in the cloud with the watermarked original images into the report, allowing managers to intuitively view the specific location and actual situation of hazards by comparing the images. This eliminates the need for repeated comparisons between text descriptions and original images, significantly improving the readability and information transmission efficiency of hazard reports and making hazard investigation results more intuitive and reliable.

[0011] Secondly, in the process of generating verifiable image and text results, the job metadata is embedded in the original job image. This job metadata forms an inseparable binding relationship between the original job image and the key information collected. The job metadata is directly integrated into the image pixels through watermark embedding. Even if the image is extracted or forwarded separately, the collected information it carries is still completely preserved. This effectively prevents the image and the collected information from being artificially separated or forged afterward, fundamentally enhancing the integrity and relevance of image evidence.

[0012] Furthermore, by invoking preset encryption rules, an anti-tampering verification code is generated based on the operation metadata and the original operation image. This verification code is then embedded in the original operation image, enabling the generated verifiable image and text results to have anti-tampering verification capabilities. In other words, the image content is bound to the operation metadata using encryption rules to generate the verification code. Any modification to the image pixels or metadata will cause the verification code to fail, thus providing a reliable guarantee of authenticity for hazard reports. This allows managers to be certain that the image evidence in the report has not been tampered with, enhancing the credibility and legal validity of hazard reports in scenarios such as safety management, accident investigation, and liability determination.

[0013] Furthermore, this application uses target glasses to acquire raw work images and transmits the images to a mobile terminal device via wireless communication. The mobile terminal device then uploads the images to a cloud server for recognition and processing, achieving separation and collaboration between the image acquisition end and the recognition and processing end. By using target glasses, it frees up the operator's hands, allowing them to complete image acquisition while wearing the glasses. This is suitable for scenarios that require two hands to operate tools or are in complex working environments such as high altitudes or narrow spaces, significantly improving the convenience and efficiency of image acquisition. At the same time, it reduces problems such as poor image quality or limited shooting angles caused by the inconvenience of operating handheld devices. Attached Figure Description

[0014] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating a typical embodiment of the hazard report generation method of this application; Figure 2 This is a schematic diagram of the hazard report generation device of this application; Figure 3 This is a schematic diagram of the structure of a computer device used in this application. Detailed Implementation

[0015] This application discloses a method for generating hazard reports, which can be programmed into a computer program product and deployed on mobile terminal devices such as mobile phones and tablets to achieve its functions. For example, in an exemplary application scenario of this application, the method can be deployed in a safety inspection application at work sites such as industrial production, construction, and power operation and maintenance. The application is installed on a mobile terminal device and runs, providing workers with functions such as image acquisition, hazard identification, watermark embedding, and report generation.

[0016] In a typical embodiment of this application, workers wear target glasses to enter the work site. These target glasses are implemented based on artificial intelligence algorithms, such as AI glasses. They are smart wearable devices equipped with a camera, display screen, and wireless communication capabilities. The target glasses' camera captures original work images, which are then sent to mobile terminal devices such as smartphones. The mobile terminal device sends the original work images to a preset cloud server, requesting the cloud server to call a preset hazard identification model to detect hazards in the original work images and obtain hazard identification results. When the mobile terminal device receives the hazard identification results returned by the cloud server, it embeds work metadata into the original work images based on a preset watermark template, generating verifiable image and text results. Finally, based on the hazard identification results and the verifiable image and text results, a target hazard report is generated.

[0017] In one embodiment, before acquiring the original job image sent by the target glasses, a watermark template selection event is responded to. Job metadata required for rendering the selected target watermark template is obtained. Off-screen rendering is performed based on the target watermark template and job metadata to generate a corresponding watermark cover bitmap. This watermark cover bitmap is then sent to the target glasses, driving the target glasses' display screen to show the watermark cover bitmap for watermark preview. Specifically, a custom view layout instruction is sent to the target glasses. This custom view layout instruction includes a container identifier and an image view identifier, driving the target glasses to initialize a watermark preview container on the display screen based on the container identifier. Then, the watermark cover bitmap is encoded into image data in a preset format, and the corresponding image data is sent to the target glasses based on the image view identifier, driving the target glasses to display the watermark cover bitmap in the watermark preview container. When a job metadata change event is responded to, the watermark cover bitmap is regenerated based on the changed job metadata and sent to the watermark preview container for update.

[0018] The hazard report generation method described in this application can be widely applied to various on-site safety inspection scenarios, such as hazard identification in factory workshops, safety inspections at construction sites, and operation and maintenance testing of power facilities. This method helps operators quickly complete on-site image acquisition and hazard identification, and by embedding tamper-proof watermarks in verifiable image and text results, it ensures the authenticity and credibility of hazard reports, effectively improving hazard identification efficiency and report quality, and reducing safety management risks caused by image tampering.

[0019] Please see Figure 1 The hazard report generation method of this application, in its typical embodiment, includes the following steps: Step S3100: Obtain the original working image sent by the target glasses; Before acquiring the original work images, communication between the mobile terminal device and the target glasses needs to be established. The target glasses have wireless communication capabilities and can establish a communication connection with the mobile terminal device via wireless communication protocols such as Bluetooth or Wi-Fi. After the connection is established, the operator can wear the target glasses to enter the work site and collect image data of the work site through the built-in camera. Once the target glasses have completed image data acquisition, the acquired image data is sent as the original work images to the mobile terminal device through the established communication connection.

[0020] In one embodiment, the target glasses and the mobile terminal device establish a control channel via Bluetooth Low Energy and a data transmission channel via Wi-Fi Direct (Wi-Fi P2P) protocol. After completing image acquisition, the target glasses first send an image acquisition signal to the mobile terminal device through the control channel. Upon receiving this signal, the mobile terminal device sends an image acquisition request to the target glasses through the Wi-Fi Direct data transmission channel. The target glasses respond to the request and transmit the original work image to the mobile terminal device through the Wi-Fi Direct channel. This dual-channel communication method effectively balances the low-latency transmission requirements of control commands with the high-bandwidth transmission requirements of image data, ensuring that image data can be stably and efficiently transmitted from the target glasses to the mobile terminal device. Metadata information of the original work image, such as shooting timestamp, geographic location information, and device identification information, can also be transmitted to the mobile terminal device simultaneously for use as a data source for subsequent watermark embedding.

[0021] In another implementation, after acquiring image data, the target glasses directly transmit the image data to the mobile terminal device via a single wireless communication channel. This single wireless communication channel can be a Bluetooth connection channel or a Wi-Fi connection channel. When using a Bluetooth connection channel, the target glasses divide the image data into multiple data packets and transmit them in packets according to the maximum transmission unit size specified by the Bluetooth protocol. The mobile terminal device receives each data packet and reassembles it to recover the complete image data. When using a Wi-Fi connection channel, the target glasses directly transmit the image data to the mobile terminal device via TCP or UDP protocols. The mobile terminal device receives, stores, and parses the data to determine it as the original working image. This single-channel communication method is suitable for application scenarios with relatively simple communication environments and low requirements for transmission rates.

[0022] Step S3200: Send the original work image to a preset cloud server, request the cloud server to call a preset hazard identification model to perform hazard detection on the original work image, and obtain the hazard identification result; An encrypted transmission channel is established between the cloud server and the mobile terminal device to ensure the data security of the original work images during transmission and prevent the image data from being intercepted or eavesdropped on by a man-in-the-middle. The mobile terminal device encapsulates the original work image into an HTTP request message and sends it to the cloud server. After receiving the original work image sent by the mobile terminal device, the cloud server calls a preset hazard identification model to detect hazards in the image. This hazard identification model is a multimodal large model or a dedicated object detection model trained based on deep learning technology. The model architecture can adopt convolutional neural networks, Transformer networks, or a hybrid architecture. During the training phase, the cloud server collects a large number of work site image samples labeled with safety hazards. These samples cover different work scenarios such as industrial production, construction, and power operation and maintenance. Hazard types include, but are not limited to, equipment malfunctions, lack of safety protection measures, violations of operating procedures, and environmental hazards. Through supervised training on the labeled samples, the hazard identification model learns to extract visual features related to safety hazards from image pixel information and establishes a mapping relationship between image content and hazard categories. During the inference phase, the cloud server inputs the original work image into the hazard identification model, and the model performs forward propagation calculations on the image and outputs the hazard identification result.

[0023] The structured information in the hazard identification results includes hazard category identifiers, which identify the specific category to which the identified safety hazard belongs, such as not wearing a safety helmet, lack of edge protection, exposed electrical equipment, and blocked fire exits. Each hazard category corresponds to a predefined category name. Additionally, the hazard identification results may include hazard location information, identifying the specific location of the hazard in the original work image in the form of bounding box coordinates or pixel masks. Bounding box coordinates are represented by the coordinates of the top left and bottom right corners of the rectangle or the center point coordinates plus width and height parameters. Pixel masks accurately outline the hazard area in the form of a binary mask image. The hazard identification results also include a hazard confidence score, a floating-point value between zero and one, representing the model's confidence in the identification results. A higher confidence score indicates that the model believes there is a greater probability of that type of hazard at that location. Furthermore, the hazard identification results may include a hazard severity level. Based on the hazard category and a predefined severity grading standard for the on-site scenario, identified hazards are divided into different levels such as general hazards, major hazards, and critical hazards, providing a basis for subsequent rectification priority ranking. Furthermore, the hazard identification results can also include hazard descriptions and corresponding rectification suggestions, which can be obtained by training a large model separately.

[0024] The cloud server encapsulates the post-processed hazard identification results into a response message and returns it to the mobile terminal device through an established encrypted transmission channel. The response message includes a request identifier, used to match the request message sent by the mobile terminal device, ensuring that the identification result correctly corresponds to the original work image. The response message may also include an identification status code to indicate whether the hazard identification was successful. Status codes include different states such as successful identification, identification failure, insufficient image quality to identify, and temporary unavailability of the model service. When identification is successful, the response message contains the complete hazard identification result. When identification fails or image quality is insufficient, the response message includes a description of the failure reason and suggested actions, such as prompting the user to retake the image or adjust the shooting angle.

[0025] Step S3300: When the hazard identification result returned by the cloud server is received, the operation metadata is embedded into the original operation image based on the preset watermark template to generate a verifiable graphic result; Upon receiving the hazard identification results from the cloud server, the watermark embedding process is triggered. Specifically, a preset watermark template is retrieved from local storage or a remote server. Alternatively, the preset watermark template can be visualized in a corresponding watermark selection control for user selection. This watermark template defines the layout structure, content composition, display style, and location information of the watermark within the original work image. The layout structure includes the relative position of the watermark area in the original work image, such as top left, bottom right, bottom center, or full image coverage. The content composition defines the information items to be included in the watermark, such as shooting timestamp, geographic location information, work identification information, and a summary of the hazard identification results. The display style defines the font, size, color, transparency, background style, and other visual presentation parameters for each information item. The location information precisely specifies the coordinate position and arrangement of each information item within the watermark area. The watermark template can be stored in a structured data format, such as JSON or XML, for easy parsing and dynamic modification.

[0026] The watermark template is parsed to determine the required job metadata. Job metadata is a set of data describing the original job image acquisition process and environment. For example, the capture timestamp precisely records the specific date and time of image capture, accurate to the millisecond level, and can be obtained from the system clock as the capture timestamp; geographic location information can be obtained through the built-in Global Positioning System (GPS) module or network positioning services, including parameters such as longitude, latitude, altitude, and positioning accuracy. When acquiring geographic location information, GPS satellite positioning is prioritized to obtain high-precision geographic coordinates; when satellite signals are poor, it automatically switches to base station positioning or wireless network positioning as a supplement. Job identification information is used to identify the personnel performing this job, and can be extracted from the currently logged-in user account information, such as employee ID, name, or user identifier. In addition, job metadata may also include auxiliary information related to the job scenario, such as job project number, job area code, weather information, and equipment serial number.

[0027] In one embodiment, after extracting the job metadata, a preset encryption rule is invoked to generate a corresponding tamper-proof verification code based on the job metadata and the original job image. This tamper-proof verification code is a security credential used to verify whether the original job image and job metadata have been tampered with during subsequent transfers. After generating the tamper-proof verification code, the job metadata and the tamper-proof verification code are embedded into the original job image according to the format and location information defined by the watermark template, ultimately generating a verifiable image and text result.

[0028] In some embodiments, the returned hazard identification results can be sent to the target glasses based on preset streaming rules, and the corresponding hazard identification text can be refreshed and displayed on the near-eye display screen of the target glasses in real time; the hazard identification text is segmented according to preset separators, and the voice broadcast interval of each segment is dynamically calculated based on the number of characters in each segment; based on the voice broadcast interval, the voice synthesis instructions of each segment of hazard identification text are sent to the target glasses in sequence, triggering the target glasses to broadcast each segment of hazard identification text in sequence.

[0029] Step S3400: Based on the hazard identification results and the verifiable graphic results, generate a target hazard report.

[0030] This target hazard report is a structured document used to comprehensively record the safety hazard investigation at the work site. It includes image evidence of the hazard, identification details, work metadata, and tamper-proof verification information, providing complete data support for subsequent safety management decisions, hazard rectification tracking, and accountability.

[0031] When generating a target hazard report, the basic framework structure of the report is first created. This framework structure defines the various components of the target hazard report and their organization. The basic framework structure includes a report header area, which displays basic information such as the report title, report number, generation time, and report version. The report header area may also include work project overview information, such as project name, work area, work type, and work start and end times; this information is extracted from the currently associated work task data. The basic framework structure also includes a hazard details area, which displays detailed information for each identified hazard, with each hazard corresponding to an independent details entry. The basic framework structure also includes a graphic evidence area, which embeds verifiable graphic results as image evidence of the hazard. The basic framework structure also includes a summary statistics area, which summarizes the overall situation of this hazard investigation, such as the total number of hazards, the distribution of the number of hazards at each severity level, and the distribution of hazard categories. The basic framework structure also includes a remediation suggestion area, which provides targeted rectification suggestions and remediation measures based on the hazard identification results. The basic framework structure may also include a verification information area, which records tamper-proof verification codes, digital signature values, and other verification credentials to facilitate subsequent verification of the report's authenticity.

[0032] After creating the report framework, the data from the hazard identification results are filled into the corresponding positions within the framework. In one embodiment, after filling in the content of each area of ​​the report, when generating the target hazard report, the operator information and approval process information are automatically associated. The operator's name, employee ID, department, contact information, and other information are extracted from the currently logged-in user account and filled into the compiler information area of ​​the hazard report.

[0033] As can be seen from the above embodiments, this application has achieved fundamental technical progress compared to traditional methods, including but not limited to: First, this application identifies potential hazards by sending the original work images to a cloud server. Upon receiving the identification results, it embeds work metadata into the original work images based on a preset watermark template to generate verifiable text and image results. Finally, it generates a target hazard report based on the hazard identification results and the verifiable text and image results, achieving deep integration of hazard identification results and original work images. This application integrates the hazard information identified in the cloud with the watermarked original images into the report, allowing managers to intuitively view the specific location and actual situation of hazards by comparing the images. This eliminates the need for repeated comparisons between text descriptions and original images, significantly improving the readability and information transmission efficiency of hazard reports and making hazard investigation results more intuitive and reliable.

[0034] Secondly, in the process of generating verifiable image and text results, the job metadata is embedded in the original job image. This job metadata forms an inseparable binding relationship between the original job image and the key information collected. The job metadata is directly integrated into the image pixels through watermark embedding. Even if the image is extracted or forwarded separately, the collected information it carries is still completely preserved. This effectively prevents the image and the collected information from being artificially separated or forged afterward, fundamentally enhancing the integrity and relevance of image evidence.

[0035] Furthermore, by invoking preset encryption rules, an anti-tampering verification code is generated based on the operation metadata and the original operation image. This verification code is then embedded in the original operation image, enabling the generated verifiable image and text results to have anti-tampering verification capabilities. In other words, the image content is bound to the operation metadata using encryption rules to generate the verification code. Any modification to the image pixels or metadata will cause the verification code to fail, thus providing a reliable guarantee of authenticity for hazard reports. This allows managers to be certain that the image evidence in the report has not been tampered with, enhancing the credibility and legal validity of hazard reports in scenarios such as safety management, accident investigation, and liability determination.

[0036] Furthermore, this application uses target glasses to acquire raw work images and transmits the images to a mobile terminal device via wireless communication. The mobile terminal device then uploads the images to a cloud server for recognition and processing, achieving separation and collaboration between the image acquisition end and the recognition and processing end. By using target glasses, it frees up the operator's hands, allowing them to complete image acquisition while wearing the glasses. This is suitable for scenarios that require two hands to operate tools or are in complex working environments such as high altitudes or narrow spaces, significantly improving the convenience and efficiency of image acquisition. At the same time, it reduces problems such as poor image quality or limited shooting angles caused by the inconvenience of operating handheld devices.

[0037] Based on any embodiment of the method in this application, before obtaining the original job image sent by the target glasses, the method includes: Step S2100: Respond to the watermark template selection event and obtain the job metadata required for rendering the selected target watermark template; Mobile terminal devices, through their running security inspection applications, provide users with an interactive interface for selecting and configuring watermark templates. Watermark template selection events can be generated by user interactions on the mobile terminal device, such as selecting a specific watermark style from the application's watermark template library through clicking, swiping, or searching. The watermark template library contains a variety of pre-defined structured watermark style templates. Each template defines the final presentation of the watermark, including but not limited to the watermark layout format, text field type, position coordinates, font style, size, color, transparency, and the style and position of graphic elements, such as company logos, security symbol placeholders, QR code frames, and various decorative border lines. Each watermark template is associated with a unique template identifier. When a user selects a watermark template, a watermark template selection event is triggered, carrying the identification information of the selected target watermark template.

[0038] In response to this event, it is necessary to obtain the job metadata required for rendering the target watermark template. The process of obtaining job metadata involves dynamically collecting and combining information from multiple data sources. First, the definition file of the selected target watermark template is parsed. This definition file can be in JSON, XML, or a similar configuration file format, which explicitly lists all the data fields that need to be filled in for rendering this template. These fields define the placeholders that need to be dynamically replaced in the watermark. Each placeholder has its corresponding metadata type identifier, and the corresponding metadata acquisition process needs to be initiated based on the type identifier of each placeholder in the template. The acquisition process can be concurrent or sequential. For example, for metadata such as shooting timestamps, the application API can be called to obtain the current precise time (UTC or local time zone), and formatted according to the predefined time format string in the template. For geographic location information, its positioning module (such as GPS, BeiDou, or network positioning based on base stations / Wi-Fi) will be activated. After obtaining the raw latitude and longitude coordinates, a reverse geocoding service can be further called to convert the coordinates into easily understandable geographic location description text, such as "XX City, XX District, XX Road, XX Number", or a specific job area name. If the template allows, latitude and longitude values ​​can also be used directly. For job identification information, the account information of the currently logged-in user can be read from local storage, and the user's employee ID, name, or role can be extracted as the job personnel identifier. Alternatively, information such as the work order number, project number, and equipment code related to the current job task can be obtained through synchronization with the backend management system. In some embodiments, the device ID can also be obtained in real time as part of the job metadata by scanning the barcode or QR code on the field equipment. Furthermore, environmental sensors can be integrated, or environmental data such as temperature, humidity, and illuminance can be obtained from the connected target glasses, and these can be embedded as watermarks as optional job metadata.

[0039] Step S2200: Perform off-screen rendering based on the target watermark template and the job metadata to generate the corresponding watermark cover bitmap; After acquiring the job metadata, the off-screen rendering process begins. The goal is to generate a watermark cover bitmap containing complete watermark information in memory without directly updating the screen display. This watermark cover bitmap is the visual layer that will ultimately be overlaid on the target glasses' display for preview. The rendering process begins with the mobile device initiating a dedicated graphics rendering context separate from the screen display. Within this context, the complete definition of the target watermark template is loaded. This definition includes not only a list of placeholders but, more importantly, detailed watermark styles, absolute or relative layout coordinates, layer order (Z-order), and drawing instructions for each element.

[0040] The rendering engine draws the layers from bottom to top, following the layer order defined in the watermark template. The first layer is the background. If the template defines a background, which could be a solid fill, a semi-transparent mask, or a pattern with a company logo, the rendering engine will draw that layer first. Then, the rendering engine iterates through all dynamic text placeholders and static graphic placeholders in the template. For each text placeholder, the rendering engine retrieves the corresponding value from the previously prepared job metadata object based on the key name, and then applies the precise style properties defined for that placeholder in the template to accurately draw the text at the specified coordinates.

[0041] In some embodiments, after all elements are drawn, the rendering engine performs compositing and post-processing, including checking for overlap between elements and ensuring proper transparency blending. The generated watermark cover bitmap exists in memory in a standard bitmap format, with its pixel size precisely matching the resolution of the target glasses preview area. For example, the bitmap may have a transparent background, meaning that only the pixel area containing the watermark elements (text, graphics) has color and transparency information, while the rest of the area is completely transparent.

[0042] Step S2300: Send the watermark cover bitmap to the target glasses and drive the display screen of the target glasses to display the watermark cover bitmap to achieve watermark preview.

[0043] Control commands and metadata are typically transmitted via a control channel established through the Bluetooth Low Energy protocol as described in the above embodiments, while media files with large data volumes, such as watermark cover bitmaps, are transmitted via a data transmission channel established through the Wi-Fi Direct protocol to ensure real-time performance.

[0044] In one embodiment, the mobile terminal device sends a custom view layout instruction to the target glasses. This instruction is a structured data packet, which may be in JSON format. The instruction contains at least two key identifiers: a container identifier, specifying the underlying container or layer on the glasses' display screen where the watermark will be carried; and an image view identifier, uniquely identifying the bitmap resource to be transmitted. This instruction is sent via a control channel. Upon receiving this instruction, the target glasses parse its content and, based on the container identifier, initialize a watermark preview container with specific size, position, and layer attributes within the graphical interface framework of the display screen. This container may be set to a transparent background and fixed at the top layer (or the second-highest layer, above the live camera view but below certain interactive controls). Initializing the container establishes a standard, addressable display area on the glasses, preventing the watermark image from floating arbitrarily on the screen.

[0045] The mobile terminal device processes the generated watermark cover bitmap. First, the device encodes the bitmap in memory to adapt for wireless transmission. According to the protocol agreed upon with the target glasses, the encoding format can be selected as PNG (supporting transparency), balancing quality and compression ratio. The encoded image data is then encapsulated into data packets. The mobile terminal device then transmits these image data packets, either streamed or all at once, to the target glasses via the data transmission channel, based on the image view identifier specified in the previous instructions.

[0046] After receiving image data, the target glasses perform decoding and display operations. In one embodiment, based on the image view identifier, the received data is located in the corresponding watermark preview container that was previously initialized. The decoder restores the data to a bitmap and then renders the bitmap into the designated watermark preview container.

[0047] Through the above embodiments, this application achieves real-time preview and dynamic updating of watermarks. Through dual-channel collaboration, the watermark cover bitmap generated on the mobile device is accurately and in real-time pushed to the smart glasses, and dynamically displayed in a preview container overlaid on the real-world view. This mechanism not only ensures that users can accurately predict the final watermark effect before shooting, achieving "preview before shooting, instant confirmation," but also supports real-time updates to the glasses' screen when the work metadata changes, ensuring strict synchronization between watermark information and the on-site work status. This effectively improves shooting quality and data entry efficiency, and optimizes the interactive experience for on-site personnel.

[0048] Based on any embodiment of the method in this application, the watermark cover bitmap is sent to the target glasses, and the display screen of the target glasses is driven to display the watermark cover bitmap to achieve watermark preview, including: Step S2310: Send a custom view layout instruction to the target glasses. The custom view layout instruction includes a container identifier and an image view identifier, and drive the target glasses to initialize a watermark preview container on the display screen based on the container identifier. A preview on the target glasses is achieved by constructing a custom view layout instruction. This instruction is a structured blueprint or layout instruction that describes the interface elements. It can be encoded using a lightweight data exchange format, such as JSON, to ensure efficient and reliable cross-platform parsing. The core data structure of the instruction should contain several key fields. Among them, the container identifier is a unique string used to identify this specific display area throughout the preview session. This identifier will be referenced in subsequent preview-related communications to ensure that operations are accurately positioned on the same view element. The image view identifier is another key field, used to associate the specific watermark bitmap resource to be sent. The instruction can also contain detailed layout parameters, whose values ​​can be specific pixel values ​​or percentages relative to the target glasses' display resolution, to precisely specify the position and size of the container. It can also include a field defining the layer level of the watermark preview container, ensuring that it appears above the live camera view but below specific prompts. In addition, style attributes such as background color, usually set to transparent, overall transparency, borders, etc., can also be included to finely control the visual appearance of the container.

[0049] The structured instruction is sent to the target glasses via an established wireless communication link. The target glasses run a client service or application to continuously listen for instructions from the paired mobile terminal device. In one embodiment, upon receiving the instruction, the glasses client starts an instruction parsing engine. This engine first verifies the format and completeness of the instruction, and then extracts the core parameters. Based on the container identifier, the target glasses check if a container with the same ID already exists in the current display. If it is the first initialization, it calls the API provided by its graphical user interface framework to dynamically create a new view control. This control applies the size, position, and style attributes defined in the instruction. This newly created view control is the watermark preview container, which is essentially a reserved, transparent, and fixed-position display window. The initialization process also includes registering the new container with the glasses' window manager and assigning it an appropriate layer to ensure that the real-time camera image is not completely obscured.

[0050] Step S2320: Encode the watermark cover bitmap into image data in a preset format, send the corresponding image data to the target glasses based on the image view identifier, and drive the target glasses to display the watermark cover bitmap in the watermark preview container; After the mobile terminal device generates a watermark cover bitmap in its memory, the watermark cover bitmap is a raw pixel matrix containing complete RGBA (red, green, blue, and transparency) channel information. In some embodiments, the watermark cover bitmap needs to be encoded to obtain image data in a preset format. After encoding, a data transmission packet is constructed, which may include a header and a body. The header contains control information, including an image view identifier to indicate which view resource slot of the target glasses the image data belongs to, while the body is filled with the encoded image data binary stream.

[0051] The constructed data packet is sent to the target glasses. Upon receiving the data packet, the target glasses first parse the packet header, extract the image view identifier, and then, based on this identifier, look up the corresponding, previously initialized watermark preview container in its internally maintained view resource mapping table. Next, the image data in the packet body is decoded. Decoding is the reverse process of encoding; the target glasses use a decoding library compatible with mobile devices to restore the binary stream to an RGB or RGBA format watermark cover bitmap. After successful decoding, the target glasses call the graphics display interface and set the obtained bitmap as the content of the target watermark preview container, that is, displaying the watermark cover bitmap in the corresponding watermark preview container.

[0052] Step S2330: In response to the job metadata change event, regenerate the watermark cover bitmap based on the changed job metadata and send it to the watermark preview container for updating.

[0053] Job metadata change events can be triggered by various user interactions. A common example is when a user manually modifies any job metadata field on a mobile device's watermark configuration interface. For instance, a user might switch operators via a dropdown menu. These events can also be triggered by preset change cycles, such as a cycle indicating that the corresponding job metadata needs to be updated every second. Furthermore, changes can also be triggered automatically by monitoring job metadata and triggering the event when a change is detected.

[0054] In response to the event, the changed metadata is extracted from it. This new data is then merged with other currently cached, unchanged job metadata to form a new, complete job metadata object, and the watermark cover bitmap is regenerated. Two different strategies can be used to optimize performance. First, for simple text field changes (such as modifying only the time or location), the target watermark template definition and graphic resources (such as icons and backgrounds) already loaded into memory can be reused. Based on the merged new metadata object, the off-screen rendering engine is invoked again. The rendering engine uses most of the cached calculation results, only rearranging and drawing the affected text placeholders. This incremental rendering method significantly reduces the computational overhead of CPU and GPU. For complex changes, such as switching watermark template sub-styles with different graphic layouts, a complete re-render may be required.

[0055] Regardless of whether incremental or full rendering is used, the regenerated watermark cover bitmap will be encoded again into the preset format. The key optimization is that it no longer requires resending the entire custom view layout instructions to initialize the container.

[0056] Through the above embodiments, when the job metadata changes, the system can quickly respond to the change event, regenerate the watermark cover bitmap through efficient incremental or full rendering, and accurately push the updated watermark image to the predetermined watermark preview container of the target glasses using the established communication channel and view identifier. This ensures that the watermark seen by the target glasses is always strictly consistent with the configuration of the mobile terminal device, and users can confirm the final watermark effect in real time, significantly improving the accuracy of data entry and the smoothness of operation in on-site operations.

[0057] Based on any embodiment of the method in this application, obtaining the original job image sent by the target glasses includes: Step S3110: Obtain the initial image captured by the target glasses; There are several triggering mechanisms for acquiring initial images from the target glasses. One mechanism is user-initiated shooting, where workers send a shooting command to the target glasses via physical buttons, touch areas, or voice commands. Upon receiving the command, the target glasses acquire images. Another mechanism is remote shooting via a mobile terminal device, which sends a remote shooting command to the target glasses through an established communication channel. Upon receiving the command, the target glasses perform image acquisition. A third mechanism is timed automatic shooting, where the target glasses or mobile terminal device are set to automatically trigger shooting at preset time intervals. For example, an image of the work site might be automatically captured every five minutes to record the temporal changes in the work process. Timed shooting is suitable for scenarios requiring continuous monitoring of the work site, such as safety inspections during long-term construction projects. The timed image sequence allows for the review of the site conditions at any given time.

[0058] After capturing the initial image, the target glasses transmit the captured image to the mobile terminal device via a communication channel.

[0059] Step S3120: Based on the preset verification rules, perform quality verification on the initial acquired image to determine whether the initial acquired image has at least one abnormal condition among motion blur, overexposure, underexposure or focus misalignment. When performing quality verification, the preset verification rules are loaded first. These verification rules include detection rules and judgment thresholds for various abnormal conditions such as motion blur, overexposure, underexposure, and focus misalignment.

[0060] For example, a gradient analysis-based detection algorithm is used for motion blur detection. Motion blur is caused by the relative motion between the camera and the scene during shooting, resulting in blurring or trailing at edges and details in the image. First, the gradient magnitude map of the initially acquired image is calculated. The gradient reflects the degree of change in pixel values ​​in the image; a clear image has high gradient values ​​in edge regions, while a blurred image has lower gradient values. After obtaining the gradient magnitude map, the overall distribution characteristics of the gradient magnitude are statistically analyzed, including the average gradient value, the standard deviation of the gradient values, and the proportion of high-gradient pixels. If the average gradient value is lower than a preset first threshold, or the proportion of high-gradient pixels is lower than a preset second threshold, the image is determined to have motion blur. A frequency domain analysis-based detection algorithm can also be used as a supplement. A Fast Fourier Transform is performed on the image to analyze its spectral distribution. The energy in the high-frequency region of a motion-blurred image is significantly lower than that of a clear image. By comparing the ratio of high-frequency energy to total energy, the degree of motion blur can be further determined.

[0061] For overexposure detection, a luminance distribution-based detection algorithm is used. Overexposure occurs when excessive exposure causes pixel values ​​in a large number of areas of the image to reach or approach their maximum values, resulting in loss of detail and a completely washed-out appearance. First, the initially acquired image is converted to a luminance color space, and luminance channel data is extracted. Then, the histogram distribution of the luminance channels is statistically analyzed, and the proportion of pixels with luminance values ​​close to their maximum values ​​is calculated. If this proportion exceeds a preset overexposure threshold, the image is determined to be overexposed.

[0062] For underexposure detection, a brightness distribution-based detection algorithm is also used. Underexposure occurs when insufficient exposure causes the overall image to be dark, resulting in the loss of detail in dark areas, which appear as a completely black image. After extracting the brightness channel data, the percentage of pixels with brightness values ​​close to the minimum value is counted. If this percentage exceeds a preset underexposure threshold, the image is determined to be underexposed.

[0063] For focusing misalignment detection, a sharpness analysis-based detection algorithm is employed. Focus misalignment occurs when the lens fails to accurately focus on the subject, resulting in a blurred image, either overall or in specific areas, making details indistinguishable. First, the image is divided into multiple local regions, such as a 3x3 grid or a finer grid. A sharpness index is calculated for each local region, using methods such as Laplacian variance or Fast Fourier Transform (FFT) high-frequency energy. The sharpness indices of each local region are compared. If the sharpness of any region is significantly lower than that of other regions, it is determined to be a local focusing misalignment; if the sharpness of all regions is lower than a preset global threshold, it is determined to be a global focusing misalignment.

[0064] After performing all anomaly detections based on the verification rules, if at least one anomaly is detected, the initially acquired image is determined to be abnormal.

[0065] Step S3130: When the condition exists, generate a corresponding retake prompt message and send it to the target glasses to display the retake prompt message on the display screen of the target glasses to guide the user to take a new picture. When the initial acquired image is determined to have at least one of the following abnormalities after quality verification: motion blur, overexposure, underexposure, or focus misalignment, a corresponding retake prompt message is generated and sent to the target glasses. The target glasses' display screen then shows the prompt message, guiding the operator to understand the quality problem of the image and retake it to obtain an original work image that meets the quality requirements.

[0066] Before generating a retake prompt, the specific anomaly type and severity information from the anomaly detection steps described above is first obtained. The anomaly type indicates which quality problems exist in the initially acquired image, such as isolated motion blur, isolated overexposure, or a combination of multiple problems such as underexposure and focus misalignment. The severity information indicates the severity level of each anomaly, such as slight, moderate, or severe.

[0067] In one embodiment, the content of the generated retake prompt information includes four parts: an anomaly type description, a problem cause analysis, improvement suggestions, and operation instructions. The anomaly description uses concise language to inform the operator of the specific problems with the image, such as motion blur (please retake the shot), overexposure (please adjust the shooting angle), underexposure (please improve lighting conditions), or inaccurate focus (please keep the image stable). The problem cause analysis briefly explains the possible causes of the anomaly, helping the operator understand the root cause. For example, motion blur may be caused by hand tremors or the target moving too fast; overexposure may be caused by shooting directly at a strong light source or a reflective surface; underexposure may be caused by backlighting or insufficient ambient light; and inaccurate focus may be caused by being too close or having an unclear focus target. Improvement suggestions provide targeted shooting adjustment solutions, such as for motion blur, suggesting holding the camera steady with both hands or using a support; for overexposure, suggesting avoiding direct light sources or adjusting the shooting angle; for underexposure, suggesting turning on a fill light or choosing a well-lit location; and for inaccurate focus, suggesting maintaining an appropriate distance or clearly focusing on the subject. The operation instructions explain the specific retake steps, such as pressing the function key to retake, speaking the retake voice command, or waiting for the automatic countdown to restart.

[0068] The generated retake prompt message is displayed on the target glasses' screen to guide the user to take another picture.

[0069] Step S3140: When it does not exist, the initial acquired image is determined as the original job image.

[0070] When the initial acquired image is determined to be free of any abnormalities such as motion blur, overexposure, underexposure, or focus misalignment after quality verification, the initial acquired image can be directly identified as the original work image for subsequent hazard identification and report generation.

[0071] Through the above embodiments, and by using preset verification rules, the initial acquired images can be quickly and objectively assessed for quality. When an anomaly is detected, a structured retake prompt containing specific problems, cause analysis, improvement suggestions, and operation guidelines is immediately generated and pushed to the target glasses' display screen in real time via the communication link, guiding on-site personnel to correct the shooting problems immediately. When the image quality is qualified, it is automatically recognized as a qualified original work image and proceeds to subsequent processes.

[0072] Based on any embodiment of the method in this application, when the hazard identification result returned by the cloud server is received, the operation metadata is embedded into the original operation image based on a preset watermark template to generate a verifiable image and text result, including: Step S3310: Parse the watermark template and determine the job metadata required for the watermark template. The job metadata includes at least one of the following: the shooting timestamp corresponding to the original job image, geographical location information, and job identification information. A watermark template is a predefined structured data file that describes how the watermark is presented in the original job image, including the watermark's layout structure, content fields, display style, and location information. Job metadata is the set of data required to populate the various content fields in the watermark template, derived from the metadata information of the original job image itself and the system information of the mobile terminal device.

[0073] The watermark template is parsed, and based on the parsed list of content fields, the required job metadata is determined. Job identification information is used to uniquely identify and associate key metadata such as on-site job tasks, operators, or job objects. Furthermore, weather conditions, temperature, etc., can be embedded as job metadata into the original job image. Specifically, the list of content fields is traversed, and the data source attribute of each field is analyzed to determine where the field value needs to be obtained. The data source attribute specifies the source of the field value, which can be embedded metadata in the original job image, system information of the mobile terminal device, query results from external services, or manual input from the user. The corresponding data acquisition process is initiated based on the different types of data sources.

[0074] For fields whose data source type is embedded metadata in the original work image, the corresponding metadata information is extracted from the original work image file. The original work images are stored in JPEG or PNG format, which supports the EXIF ​​metadata standard. The EXIF ​​data segment of the original work image is parsed to extract various shooting parameters and scene information, including shooting timestamps or geographic location information. For fields whose data source type is mobile terminal device system information, the corresponding system interface can be called to obtain the data. The system time field obtains the current date and time through the time API provided by the operating system. For fields whose data source type is external service query results, data can be obtained by calling third-party service interfaces. The weather information field calls the meteorological service API, inputs the current geographic location coordinates, and queries to obtain local weather conditions, temperature, humidity, wind speed, visibility, and other meteorological data. For fields whose data source type is user-inputted, an input control can be displayed on the user interface, waiting for the user to input the value of the field. The remarks information field can be manually entered; operators can enter special instructions, precautions, or descriptions of problems discovered during the shooting. Custom number fields can also be manually entered, such as the equipment inspection serial number or the work order number for hazard rectification.

[0075] Step S3320: Invoke the preset encryption rules and generate a corresponding tamper-proof verification code based on the job metadata and the original job image; In one embodiment, the various information items in the job metadata are first concatenated into a string in a predetermined order. The concatenation order is predefined by encryption rules, such as concatenating the shooting timestamp, geographical location information, and job identification information sequentially, with each information item separated by a specific delimiter. Then, the pixel data of the original job image is read, and a hash calculation is performed on the image's pixel matrix to generate an image hash value. Commonly used hash algorithms include the Secure Hash Algorithm series or the Message Digest Algorithm series. These algorithms can map image data of arbitrary length to a fixed-length hash value and are highly sensitive to minor changes in the input data; even a tiny modification to the image data will significantly change the generated hash value. The concatenated job metadata string is then concatenated with the image hash value to form the data to be encrypted. Subsequently, a preset encryption key is used to perform encryption operations on the data to be encrypted, generating a tamper-proof verification code. The encryption algorithm can be a symmetric encryption algorithm, such as the Advanced Encryption Standard (AES), which is fast and suitable for real-time computation on mobile terminal devices; or an asymmetric encryption algorithm, such as public-key encryption, which offers higher security but has a relatively higher computational cost. The generated tamper-proof verification code is a fixed-length character sequence with high randomness and unpredictability. Any unauthorized tampering will cause the verification code to fail.

[0076] Step S3330: Embed the job metadata and the anti-tampering verification code into the original job image according to the format and location information defined by the watermark template to generate the verifiable image and text result.

[0077] The tamper-proof verification code and corresponding job metadata generated in the preceding steps are embedded into the original job image according to the format and location information defined in the watermark template. In one embodiment, firstly, according to the display style parameters defined in the watermark template, each job metadata element is rendered into a visual text or graphic element. Then, the rendered watermark element is superimposed onto the original job image according to the location information defined in the watermark template. The superimposition method can be a transparency blending method, where the watermark element covers the corresponding area of ​​the original image with a certain transparency, making the watermark information visible without severely obscuring the content of the original image. The transparency value is controlled by the transparency parameter in the watermark template. The superimposition method can also be a background frame method, where a semi-transparent background rectangle is drawn below the watermark text element to improve the readability of the watermark text on a complex image background. The color and transparency of the background frame are defined by the watermark template. The superimposition method can also be an edge blending method, where the watermark element is gradually blended into the edge area of ​​the original image, allowing the watermark to naturally integrate into the image boundary and reducing visual interference with the main content of the image. During the overlay process, it is also necessary to address the conflict between the watermark element and important visual content in the original image. For example, if the watermark area happens to cover the location of the hazard identified by the hazard identification result, the embedding position of the watermark can be automatically adjusted, or the watermark can be split into multiple small blocks and embedded into different edge areas of the image to ensure that the visibility of the hazard area is not affected.

[0078] Through the above embodiments, watermark template parsing enables intelligent identification and collection of diverse operational metadata, providing rich and accurate context for the watermark. Based on this, an anti-tampering verification code is generated by combining operational metadata with the original image hash using encryption rules, providing cryptographic-level protection for the integrity and authenticity of the final image. Finally, according to the style and position defined in the template, watermark elements, metadata, and verification codes are intelligently embedded into the image in diverse overlay methods. This ensures that key information in the original image is not obscured, generating a verifiable image and text result that contains detailed operational background and possesses high credibility. This ensures the authenticity, accuracy, and traceability of each on-site image, providing a reliable data foundation for subsequent hazard identification, report generation, and accountability.

[0079] Based on any embodiment of the method in this application, obtaining the original job image sent by the target glasses includes: Step S3150: Receive the continuous image sequence acquired by the target glasses, the continuous image sequence including multiple image frames; This step acquires a continuous stream of dynamic visual information from the target glasses, rather than a single static image. When the operator interacts with the glasses through specific actions (such as pressing and holding the camera button on the target glasses, issuing a voice command to "start recording," or triggering continuous capture mode on a mobile app), the built-in camera of the target glasses is activated and enters continuous capture mode. In this mode, the camera continuously captures the scene at a preset frame rate, generating a series of image data arranged chronologically, i.e., a continuous image sequence, where each image is called an image frame.

[0080] In some embodiments, the continuous image sequence may be formed by taking pictures of the same work site from different angles within a preset time period and arranging the pictures in chronological order.

[0081] The target glasses assign a precise timestamp (accurate to milliseconds) and frame number to each frame in a continuous image sequence. Then, the target glasses continuously transmit the data packets of the continuous image sequence to the mobile terminal device via an established high-speed data channel using a streaming media transmission protocol.

[0082] Step S3160: Based on the preset quality detection rules, perform quality scoring on each image frame in the continuous image sequence, and determine the target image frame whose quality score meets the preset threshold. Upon receiving a continuous image sequence consisting of multiple image frames, the system first extracts the visual features of each frame and calculates multiple quality scoring metrics. Common preset detection rules include sharpness scoring, which quantifies the sharpness and detail richness of an image by calculating its gradient energy (such as the Tenengrad gradient function), Laplacian operator variance, or frequency-domain energy concentration; a higher score indicates greater sharpness and a lower probability of motion blur. It also includes exposure evaluation scoring, which assesses the overall exposure level by analyzing the image's brightness histogram. The process involves calculating the average brightness value of the image and comparing it to the ideal mid-gray brightness value to obtain a deviation score. Simultaneously, it evaluates the pixel stacking in the histogram of extreme highlights and shadows, deducting points for overexposure (blowout highlights) and underexposure (completely black shadows), ultimately providing a comprehensive score for whether the exposure is appropriate. It also includes a contrast score, calculating the global or local contrast of the image (e.g., using standard deviation or the Michelson contrast formula). Appropriate contrast helps highlight the difference between the target object and the background; a low score may indicate a hazy image with unclear details. Noise level assessment is also included, estimating the intensity of image noise by analyzing pixel value fluctuations in flat areas of the image (e.g., calculating local variance). Images with excessive noise receive lower scores. Furthermore, it may include a focus area assessment. If the glasses camera provides focus area information, or if the algorithm detects potential regions of interest in the image (e.g., through saliency detection or object detection boxes), the sharpness score of these key areas is specifically calculated to ensure that the sharpness of the focus is more important than the sharpness of the background.

[0083] For each frame, the aforementioned metrics are calculated separately and converted into normalized scores (e.g., from 0 to 100). Then, based on preset weights, these scores are weighted and synthesized into a final comprehensive quality score, which represents the overall quality level of the image frame under preset quality detection rules. By traversing the comprehensive quality scores of all frames in the entire continuous image sequence and setting a preset quality threshold, image frames with comprehensive quality scores greater than the threshold are marked as target image frames.

[0084] Step S3170: Extract the target image frame from the continuous image sequence to form the original job image.

[0085] The target image frames marked in the aforementioned steps are extracted from the continuous image sequence to form the original work image. Subsequently, watermarks can be embedded into each image frame in the continuous image sequence to generate corresponding hazard reports. Specifically, the trained hazard identification model can identify hazards in each target image frame in the continuous image sequence and generate corresponding hazard identification results for different target image frames. At this point, watermarks can be embedded into each image frame individually, or the watermark for the continuous image sequence can be embedded into one image frame, depending on the specific situation.

[0086] Through the above embodiments, a continuous image sequence containing multiple angles and with a time sequence is acquired using target glasses, providing more comprehensive visual coverage of the scene. Subsequently, quality detection rules are applied to score the quality of each frame in the continuous image sequence, and target image frames that meet the quality standards are selected. This embodiment effectively solves the problem that a single shot may not be able to fully and clearly capture hazard features in complex scenes (such as equipment obstruction, uneven lighting, and limited viewing angles). By selecting multiple frames from the continuous sequence, more representative input is provided to the cloud-based hazard identification model, thereby significantly improving the accuracy and detection rate of hazard identification and reducing the risk of missed identification due to single-frame image quality or viewing angle issues.

[0087] Based on any embodiment of the method in this application, after generating the target hazard report, the following steps are included: Step S4100: Receive a verification request for the verifiable graphic and textual results in the target hazard report, and extract the operation metadata and anti-tampering verification code embedded in the verifiable graphic and textual results; This step is designed to respond to authenticity verification requests initiated externally (e.g., by management, auditing systems) or internally. It assumes a target hazard report (which can be a PDF, Word document, or webpage report) has been generated and distributed to relevant parties. This report contains verifiable visual evidence—images embedded with watermarks—as key evidence. When any party needs to verify the authenticity and completeness of this visual evidence, a verification request will be initiated. This request can be triggered in various ways. For example, in a report viewing system, a user can click a "verify authenticity" button next to the image; or, an automated auditing program can periodically scan the report library and initiate batch verification requests for all images. This verification request is sent to a designated verification server endpoint, which can be deployed on a mobile device or on a centralized, publicly accessible verification server.

[0088] Upon receiving a request, the verification service endpoint (hereinafter referred to as the "verification service") first needs to locate and retrieve the image data corresponding to the verifiable image and text result to be verified from the request. The request may contain a reference to the image storage location (such as a URL or file path). The verification service then loads the image data.

[0089] Then, the previously embedded job metadata and anti-tampering verification code are parsed from the image data. The extraction process is the reverse of the aforementioned step "embedding the job metadata and the anti-tampering verification code into the original job image according to the format and position information defined by the watermark template to generate the verifiable image and text result," in order to extract the job metadata and anti-tampering verification code. After successful extraction, the extracted data is used as the basis for subsequent cryptographic comparison and verification of whether the image has not been tampered with since its generation.

[0090] Step S4200: Based on the job metadata and the original job image, regenerate the verification code by calling the preset encryption rules; After extracting the job metadata and the tamper-proof verification code, the exact same cryptographic operations as when generating the tamper-proof verification code need to be performed to determine if the data is consistent. Specifically, the same encryption rules are applied to the job metadata extracted from the verifiable image and text results to obtain the verification code.

[0091] Step S4300: Compare the anti-tampering verification code with the verification code. If they match, it is determined that the verifiable image and text result has not been tampered with. If they do not match, a tampering alarm message is generated.

[0092] The verification code calculated in the previous steps is compared with the anti-tampering verification code extracted from the verifiable image and text result. If the two are the same, it can be determined that the verifiable image and text result has not been tampered with. If the two are different, the verification is determined to be unsuccessful, and the corresponding tampering alarm information is triggered.

[0093] Through the above embodiments, when a verification request is initiated for a generated hazard report, the system can automatically receive and locate the image to be verified, which is embedded with a watermark and an anti-tampering verification code. By executing reverse parsing and encryption rules that are strictly consistent with those used during generation, the system extracts the work metadata and the original anti-tampering verification code from the image and independently regenerates the verification code. By cryptographically comparing the two, it can accurately determine whether the verifiable image and text result has been tampered with since its generation. This embodiment provides a reliable third-party verification capability for evidence in hazard reports, ensuring that the authenticity and integrity of the hazard report as legal or administrative evidence can always be traced and verified during the distribution, circulation, and archiving process, thus consolidating anti-counterfeiting credibility and evidentiary validity from the end of the process.

[0094] Please see Figure 2This invention provides a hazard report generation device to meet one of the purposes of this application. It is a functional embodiment of the hazard report generation method of this application. The device includes an image acquisition module 3100, a hazard identification module 3200, a watermark embedding module 3300, and a report generation module 3400. The image acquisition module 3100 is configured to acquire the original work image sent by the target glasses. The hazard identification module 3200 is configured to send the original work image to a preset cloud server, requesting the cloud server to call a preset hazard identification model to perform hazard detection on the original work image and obtain a hazard identification result. The watermark embedding module 3300 is configured to, upon receiving the hazard identification result returned by the cloud server, embed work metadata into the original work image based on a preset watermark template to generate a verifiable graphic result. The report generation module 3400 is configured to generate a target hazard report based on the hazard identification result and the verifiable graphic result.

[0095] Based on any embodiment of the device in this application, before the image acquisition module 3100, the device includes: a metadata acquisition module, configured to respond to a watermark template selection event and acquire the job metadata required for rendering the selected target watermark template; a bitmap generation module, configured to perform off-screen rendering processing based on the target watermark template and the job metadata to generate a corresponding watermark cover bitmap; and a watermark preview module, configured to send the watermark cover bitmap to the target glasses and drive the display screen of the target glasses to display the watermark cover bitmap to achieve watermark preview.

[0096] Based on any embodiment of the device in this application, the watermark preview module includes: a container initialization module, configured to send a custom view layout instruction to the target glasses, the custom view layout instruction including a container identifier and an image view identifier, driving the target glasses to initialize a watermark preview container on the display screen based on the container identifier; a bitmap display module, configured to encode the watermark cover bitmap into image data of a preset format, send the corresponding image data to the target glasses based on the image view identifier, and drive the target glasses to display the watermark cover bitmap in the watermark preview container; and a bitmap update module, configured to respond to the job metadata change event, regenerate the watermark cover bitmap based on the changed job metadata, and send it to the watermark preview container for update.

[0097] Based on any embodiment of the device in this application, the image acquisition module 3100 includes: an initial image acquisition module, configured to acquire an initial image captured by the target glasses; a quality verification module, configured to perform quality verification on the initial image based on preset verification rules, and determine whether the initial image has at least one abnormal condition among motion blur, overexposure, underexposure, or focus misalignment; a prompt information generation module, configured to generate corresponding retake prompt information and send it to the target glasses when such an abnormal condition exists, so as to display the retake prompt information on the display screen of the target glasses and guide the user to retake the shot; and an image determination module, configured to determine the initial image as the original working image when such an abnormal condition does not exist.

[0098] Based on any embodiment of the device in this application, the watermark embedding module 3300 includes: a template parsing module, configured to parse the watermark template and determine the job metadata required for the watermark template, wherein the job metadata includes at least one of the shooting timestamp, geographical location information, and job identification information corresponding to the original job image; a verification code generation module, configured to call a preset encryption rule and generate a corresponding anti-tampering verification code based on the job metadata and the original job image; and an image and text result generation module, configured to embed the job metadata and the anti-tampering verification code into the original job image according to the format and location information defined by the watermark template, and generate the verifiable image and text result.

[0099] Based on any embodiment of the device in this application, the image acquisition module 3100 further includes: an image sequence receiving module, configured to receive a continuous image sequence acquired by the target glasses, the continuous image sequence including multiple image frames; a quality scoring module, configured to perform quality scoring on each image frame in the continuous image sequence based on preset quality detection rules, and determine a target image frame whose quality score meets a preset threshold; and an image frame extraction module, configured to extract the target image frame from the continuous image sequence to form the original working image.

[0100] Based on any embodiment of the device in this application, after the report generation module 3400, it includes: a verification request module, configured to receive a verification request for the verifiable graphic results in the target hazard report, and extract the operation metadata and anti-tampering verification code embedded in the verifiable graphic results; a verification code generation module, configured to regenerate the verification code based on the operation metadata and the original operation image by calling a preset encryption rule; and a verification code comparison module, configured to compare the anti-tampering verification code with the verification code, and if they match, determine that the verifiable graphic results have not been tampered with, and if they do not match, generate a tampering alarm message.

[0101] To address the aforementioned technical problems, embodiments of this application also provide a computer device. For example... Figure 3 The diagram shows the internal structure of a computer device. This computer device includes a processor, a computer-readable storage medium, a memory, a network interface, and various communication components connected via a system bus. The computer-readable storage medium stores an operating system, a database, and computer-readable instructions. The database may store control information sequences. When the computer-readable instructions are executed by the processor, they enable the processor to implement a hazard report generation method. The processor of this computer device provides computing and control capabilities, supporting the operation of the entire computer device. The memory of this computer device may store computer-readable instructions. When these computer-readable instructions are executed by the processor, they enable the processor to execute the hazard report generation method of this application. The network interface of this computer device is used for communication with a terminal. Those skilled in the art will understand that… Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0102] In this embodiment, the processor is used to execute... Figure 2 The system contains the specific functions of each module and its sub-modules, and the memory stores the program code and various data required to execute these modules or sub-modules. The network interface is used for data transmission between the user terminal and the server. In this embodiment, the memory stores the program code and data required to execute all modules / sub-modules in the hazard report generation device of this application, and the server can call the server's program code and data to execute the functions of all sub-modules.

[0103] This application also provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the vulnerability report generation method of any embodiment of this application.

[0104] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. This computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0105] Those skilled in the art will understand that the steps, measures, and solutions in the various operations, methods, and processes discussed in this application can be alternated, modified, combined, or deleted. Furthermore, other steps, measures, and solutions in the various operations, methods, and processes discussed in this application can also be alternated, modified, rearranged, decomposed, combined, or deleted. Furthermore, steps, measures, and solutions in the prior art that are similar to those in the open-source operations, methods, and processes of this application can also be alternated, modified, rearranged, decomposed, combined, or deleted.

[0106] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for generating hazard reports, characterized in that, include: Acquire the raw image of the task sent by the target glasses; The original work image is sent to a preset cloud server, which requests the cloud server to call a preset hazard identification model to perform hazard detection on the original work image and obtain the hazard identification result. When the hazard identification result returned by the cloud server is received, the operation metadata is embedded into the original operation image based on the preset watermark template to generate a verifiable graphic result; Based on the hazard identification results and the verifiable graphic results, a target hazard report is generated.

2. The method for generating a hazard report according to claim 1, characterized in that, Before acquiring the raw job image sent by the target glasses, the following steps are included: Respond to the watermark template selection event to obtain the job metadata required for rendering the selected target watermark template; Off-screen rendering is performed based on the target watermark template and the job metadata to generate a corresponding watermark cover bitmap. The watermark cover bitmap is sent to the target glasses, and the display screen of the target glasses is driven to display the watermark cover bitmap to achieve watermark preview.

3. The method for generating a hazard report according to claim 2, characterized in that, Sending the watermark cover bitmap to the target glasses and driving the target glasses' display screen to show the watermark cover bitmap to achieve watermark preview includes: Send a custom view layout instruction to the target glasses. The custom view layout instruction includes a container identifier and an image view identifier. Drive the target glasses to initialize a watermark preview container on the display screen based on the container identifier. The watermark cover bitmap is encoded into image data in a preset format, and the corresponding image data is sent to the target glasses based on the image view identifier, driving the target glasses to display the watermark cover bitmap in the watermark preview container; In response to the job metadata change event, a new watermark cover bitmap is generated based on the changed job metadata and sent to the watermark preview container for updating.

4. The method for generating a hazard report according to claim 1, characterized in that, Acquire the raw job images sent by the target glasses, including: Acquire the initial image captured by the target glasses; Based on preset verification rules, the quality of the initial acquired image is verified to determine whether the initial acquired image has at least one abnormal condition, such as motion blur, overexposure, underexposure, or focus misalignment. When the condition is met, a corresponding retake prompt message is generated and sent to the target glasses to display the retake prompt message on the target glasses' screen, guiding the user to take the photo again; If it does not exist, the initial acquired image is determined as the original job image.

5. The method for generating a hazard report according to claim 1, characterized in that, Upon receiving the hazard identification result returned by the cloud server, based on a preset watermark template, the operation metadata is embedded into the original operation image to generate a verifiable image and text result, including: The watermark template is parsed to determine the job metadata required for the watermark template. The job metadata includes at least one of the shooting timestamp, geographic location information, and job identification information corresponding to the original job image. The preset encryption rules are invoked to generate a corresponding tamper-proof verification code based on the job metadata and the original job image; The job metadata and the anti-tampering verification code are embedded into the original job image according to the format and location information defined by the watermark template to generate the verifiable image and text result.

6. The method for generating a hazard report according to claim 1, characterized in that, Acquiring the raw job image sent by the target glasses also includes: Receive a continuous image sequence acquired by the target glasses, the continuous image sequence including multiple image frames; Based on preset quality detection rules, each image frame in the continuous image sequence is scored to determine the target image frame whose quality score meets a preset threshold. The target image frame is extracted from the continuous image sequence to form the original job image.

7. The method for generating a hazard report according to claim 1, characterized in that, After generating the target hazard report, the following steps are included: Receive a verification request for the verifiable graphic and textual results in the target hazard report, and extract the operation metadata and anti-tampering verification code embedded in the verifiable graphic and textual results; Based on the job metadata and the original job image, the verification code is regenerated by calling the preset encryption rules; The anti-tampering verification code is compared with the verification code. If they match, it is determined that the verifiable image and text result has not been tampered with. If they do not match, a tampering alarm message is generated.

8. A hazard report generation device, characterized in that, include: The image acquisition module is configured to acquire the raw job image sent by the target glasses; The hazard identification module is configured to send the original work image to a preset cloud server, request the cloud server to call a preset hazard identification model to perform hazard detection on the original work image, and obtain the hazard identification result; The watermark embedding module is configured to embed the operation metadata into the original operation image based on a preset watermark template when the hazard identification result returned by the cloud server is received, thereby generating a verifiable image and text result. The report generation module is configured to generate a target hazard report based on the hazard identification results and the verifiable graphic results.

9. A computer device comprising a central processing unit and a memory, characterized in that, The central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implemented according to any one of claims 1 to 7, which, when invoked by a computer, executes the steps included in the corresponding method.

Images