Virtual-real combined interactive touring method and system for digital museum

Abstract

The invention discloses a virtual and real combined interactive touring method and system for a digital museum. The method comprises the following steps: (S1) content preprocessing and safety management; (S2) pre-judging and pre-loading user behaviors; (S3) carrying out computing power adaptation and rendering; (S4) carrying out AR interaction personalized adjustment; and (S5) cache and system collaboration. According to the virtual and real combined interactive sightseeing method and system for the digital museum, through position pre-judgment, block loading and AI pre-rendering, core experience loading time of high-precision content is compressed within 3 seconds, full content loading is smaller than or equal to 5 seconds, and the problem that the high-precision content is stuck is solved; through block chain fragmentation and dual verification, the content tampering recognition rate is 100%, and the authenticity of high-precision content is ensured; through AI pre-rendering and LOD dynamic adaptation, the real-time computing power occupancy of the equipment is reduced to be within 30%, and the low-computing-power equipment can also reach a smooth frame rate of more than 30fps.

Description

Technical Field

[0001] This application relates to the field of AR experience technology, and in particular to a method and system for interactive virtual and real-world tours of digital museums. Background Technology

[0002] Traditional museum exhibitions often suffer from insufficient explanatory information and poor interactivity. Visitors increasingly demand a greater sense of participation and experiential engagement with exhibition content. AR technology can enhance visitor participation and learning interest by providing dynamic information displays and virtual interactions.

[0003] Chinese invention patent CN119027623B discloses a method and system for interactive AR experience of museum exhibits, relating to the field of interactive AR experience technology. During exhibit display adjustment, by incorporating the cultural background information of the exhibits into the design of the AR display screen, users can better understand the historical background and cultural value of each exhibit. During display screen analysis, by adjusting the display screen information of the AR device in real time, optimization can be made according to the visual comfort and preferences of each visitor, ensuring that each visitor can obtain the best visual experience in different environments and avoiding visual fatigue or discomfort caused by display discomfort. During voice broadcast analysis, by analyzing environmental noise and visitor voice input, the system can dynamically adjust the speech rate and volume of the voice broadcast, ensuring that the voice information remains clear and audible under various environmental noise conditions, improving the effectiveness of interaction and user experience.

[0004] However, the above technical solutions still have the following problems: they only adjust the AR interaction experience based on the user's physiological and environmental parameters, but cannot load, store, or securely store high-precision content; when loading complex models, the conventional full loading method will result in a loading time of more than 10 minutes, requiring users to wait for a long time and destroying the immersion; and they rely too much on the device's real-time calculation; when browsing multiple exhibits, the storage usage is too high, which may lead to loading failure. Summary of the Invention

[0005] This application aims to at least partially solve one of the technical problems in the aforementioned technologies.

[0006] To achieve the above objectives, the first aspect of this application proposes a method for interactive virtual and real-world tours of digital museums, comprising the following steps:

[0007] (S1) Content Preprocessing and Security Management: Cultural backgrounds are divided according to the origin period of exhibits and historical events, and unique codes are assigned. A two-dimensional block sorting algorithm based on cultural relevance and visual weight is used to determine the block priority. AR content is split into texture layer, 3D model layer, and interactive information layer. Then, a unique hash value for each block is generated using the SHA-256 hash algorithm. The block ID and cultural background code are combined and packaged into a security verification package. A consortium blockchain is built using the Hyperledger Fabric 2.4 framework, and the PBFT consensus algorithm is used to put the verification package on the blockchain to ensure the integrity of the content.

[0008] (S2) User behavior prediction and preloading: Integrate the traffic data of the exhibition area with the positioning, gyroscope and gaze data of AR devices, construct a two-factor prediction model through logistic regression, and combine the vector angle algorithm to determine the deviation angle. When the prediction probability is ≥0.7 or the dwell time is ≥prediction threshold, preloading is triggered. Multi-threaded concurrent loading is adopted. When the user deviates, the non-core block loading is interrupted and resources are released.

[0009] (S3) Computing power adaptation and rendering: The computing power level of the device is divided by OpenGL ES parameters and Geekbench score. The LOD dynamic adaptation algorithm is used to match the corresponding precision 3D model and texture. The lightweight Diffusion pre-rendering algorithm is deployed at the edge node to pre-render high-frequency viewpoints. Combined with the deferred rendering algorithm, the 3D model rendering is processed in three stages to ensure the frame rate is ≥30fps.

[0010] (S4) Personalized Adjustment of AR Interaction: Collect the pupil diameter di, current ambient light intensity li, and light source color temperature Ti of the i-th visitor, and calculate the comprehensive display evaluation index αi using a weighted summation algorithm. The formula is as follows:

[0011]

[0012] Where d′, l′, and T′ represent the standard pupil diameter, standard illumination intensity, and standard color temperature, respectively, and κ1=0.4, κ2=0.3, and κ3=0.3, the AR image parameters are dynamically adjusted based on αi; the speech rate vi, sound decibel Bri, and environmental noise decibel Bit at the t-th acquisition time point are collected for the i-th visitor, and the speech adjustment evaluation coefficient βit is calculated using a weighted summation algorithm, the formula of which is:

[0013]

[0014] Where v′, Br′, and B′ represent standard speech rate, standard sound decibel, and standard noise decibel, respectively, and λ1=0.45, λ2=0.3, and λ3=0.25, the speech rate and volume are dynamically adjusted based on βit. When the user browses for more than 4 seconds, the ORB feature detection algorithm is used to match the AR image and text description and display them in a floating manner.

[0015] (S5) Cache and system collaboration: The LRU cache eviction algorithm is used to evict idle blocks, while the gRPC protocol is used to realize module interaction. An adaptive exception adjustment algorithm is used to handle loading timeouts, insufficient computing power, and network interruptions. The InfluxDB time series data query algorithm is used to store and query system operation indicators.

[0016] In addition, the virtual-real interactive tour method for digital museums proposed above in this application may also have the following additional technical features:

[0017] As a further description of the above technical solution:

[0018] In step (S1), the texture layer is divided into 24 512×512 pixel blocks according to 8K resolution, the 3D model layer is divided into 10 blocks of 200,000-300,000 faces, and the interactive information layer is divided into 5 JSON format blocks.

[0019] As a further description of the above technical solution:

[0020] In step (S1), the consortium blockchain includes one museum master node, one edge node, and one technical service provider verification node. The PBFT consensus requires ≥2 / 3 of the nodes to reach an agreement, the block generation time is ≤2 seconds, and if the local hash verification is inconsistent during block loading, it will be retransmitted via HTTPS.

[0021] As a further description of the above technical solution:

[0022] In step (S2), the training data for the two-factor prediction model consists of pedestrian flow data and AR device data from the past three months. The model accuracy is ≥85%, and the calculation time for the vector angle algorithm is ≤30 milliseconds.

[0023] As a further description of the above technical solution:

[0024] In step (S3), the LOD adaptation standard is as follows: high computing power devices load a 1.2 million-face model and 8K texture; medium computing power devices load a 600,000-face model and 4K texture; and low computing power devices load a 250,000-face model and 2K texture.

[0025] As a further description of the above technical solution:

[0026] The standard parameter values ​​in step (S4) are:

[0027]

[0028] The evaluation index is shown. Calculation time ≤ 30 milliseconds, screen parameter adjustment response ≤ 100 milliseconds, voice adjustment evaluation coefficient. Calculation time ≤ 50 milliseconds, voice parameter adjustment delay ≤ 200 milliseconds.

[0029] As a further description of the above technical solution:

[0030] In step (S4), the ORB feature detection algorithm extracts ≥500 feature points / frame, the Hamming distance matching threshold is ≤50, the comparison time is ≤150 milliseconds, the size of the text floating label is 300×150 pixels, it automatically avoids the main body of the product, and the click response is ≤100 milliseconds.

[0031] As a further description of the above technical solution:

[0032] In step (S5), the LRU cache eviction rate is ≥100MB / s, the total storage usage for browsing multiple exhibits is ≤500MB, the adaptive anomaly adjustment strategy is to switch to low-precision backup content when loading timeout, and to gradually reduce the number of model faces by 10% when the frame rate is <25fps, and to enable the local core area cache when the network is interrupted.

[0033] The second aspect of this application proposes a virtual-real integrated interactive tour system for digital museums, including:

[0034] The content preprocessing and security module divides and encodes high-precision AR content based on cultural background, and then performs two-dimensional block sorting, SHA-256 hash calculation, and PBFT consensus on the Hyperledger Fabric 2.4 framework for on-chain processing.

[0035] The user prediction and preloading module collects data on pedestrian flow and AR devices, and runs a two-factor prediction model and a vector angle algorithm. The formula for the vector angle algorithm is:

[0036] This triggers preloading and executes multi-threaded loading;

[0037] The computing power adaptation rendering module divides the device computing power levels and performs LOD adaptation, lightweight Diffusion pre-rendering and deferred rendering.

[0038] The interactive personalization adjustment module collects data from the i-th visitor. respectively through formula , calculate

[0039] Then, based on the calculation results, the display / voice parameters are adjusted, ORB detection is performed, and floating text descriptions are displayed.

[0040] The caching and coordination module performs LRU cache eviction, enables module interaction via gRPC, and runs adaptive exception handling and time-series data query algorithms.

[0041] As a further description of the above technical solution:

[0042] The computing power adaptation rendering module includes a computing power detection unit, a pre-rendering unit, and a deferred rendering unit, while the caching and collaboration module includes a data monitoring unit.

[0043] Advantages of this invention:

[0044] According to the virtual-real interactive tour method and system for digital museums proposed in this application, the loading time of the core experience of high-definition content is compressed to within 3 seconds and the loading time of the full content is ≤5 seconds through location prediction, segmented loading and AI pre-rendering, thus solving the problem of lag in high-definition content.

[0045] By using blockchain sharding and dual verification, a 100% content tampering detection rate is achieved, ensuring the authenticity of highly detailed content;

[0046] By using AI pre-rendering and LOD dynamic adaptation, the real-time computing power usage of the device is reduced to less than 30%, and even low-computing-power devices can achieve a smooth frame rate of more than 30fps.

[0047] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0048] 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:

[0049] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0050] Figure 2 This is a system principle block diagram of the present invention;

[0051] Figure 3 This is a schematic diagram of the process logic of the present invention. Detailed Implementation

[0052] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0053] The following describes, with reference to the accompanying drawings, a virtual-real integrated interactive tour method and system for digital museums according to embodiments of this application.

[0054] like Figure 1 As shown in Embodiment 1 of this application, the method for interactive virtual and real-world tours of digital museums includes the following steps:

[0055] Content preprocessing and security management: Cultural backgrounds are categorized according to the origin period and historical events of exhibits and assigned unique codes (format: historical stage-event type-exhibit ID). A two-dimensional block sorting algorithm based on cultural relevance (weight 0.4) and visual weight (weight 0.6) is used to determine block priority. AR content is split into texture layer, 3D model layer, and interactive information layer. Then, a unique hash value for each block is generated using the SHA-256 hash algorithm. Combined with the block ID and cultural background code, the blocks are packaged into a security verification package. A consortium blockchain is built using the Hyperledger Fabric 2.4 framework, and the PBFT consensus algorithm is used to upload the verification package to the blockchain to ensure content integrity.

[0056] In the above steps, the AR content splitting parameters are as follows: the texture layer is split into 24 512×512 pixel blocks at 8K resolution (using ETC2 format, 45-55MB / block), the 3D model layer is split into 10 blocks of 200,000-300,000 polygons (using glTF 2.0 format, 18-22MB / block, total number of polygons ≥ 250,000), and the interactive information layer is split into 5 JSON format blocks (5-7MB / block).

[0057] In the above steps, the consortium blockchain configuration includes 1 museum master node, 2-3 edge nodes per exhibition area, and 1 technical service provider verification node. PBFT consensus requires ≥2 / 3 of the nodes to reach an agreement. The block generation time is ≤2 seconds. If the local hash verification is inconsistent during block loading, it will be retransmitted via HTTPS (with ≤3 retries at 100 millisecond intervals).

[0058] User behavior prediction and preloading: Integrating exhibition area pedestrian flow data with AR device positioning, gyroscope, and gaze data, a two-factor prediction model is constructed through logistic regression, combined with a vector angle algorithm, the formula of which is: Calculate the angle of deviation between the line of sight and the target exhibit. User gaze vector, The target exhibit's position vector is used to determine the deviation angle. Preloading is triggered when the error is ≤ ±5°, the prediction probability is ≥ 0.7, or the dwell time is ≥ the prediction threshold (where the average movement time from the adjacent area to the target exhibit is × 1.8). Multi-threaded concurrent loading is used (the maximum number of threads is the number of CPU cores on the device / 2). When the user deviates, the non-core block loading is interrupted and resources are released.

[0059] In the above steps, the training data for the two-factor prediction model consists of pedestrian flow data from the past 3 months (5 minutes / batch, sample size ≥1000 items / day) and AR device data (positioning sampling at 1Hz, gyroscope sampling at 10Hz). At this time, the model accuracy is ≥85%, and the calculation time using the vector angle algorithm is ≤30 milliseconds.

[0060] Computing power adaptation and rendering: Device computing power levels are divided by OpenGL ES parameters and Geekbench scores. The LOD dynamic adaptation algorithm is used to match the corresponding precision 3D models and textures. The lightweight Diffusion pre-rendering algorithm (using the Stable Diffusion XL Tiny fine-tuning model) is deployed at the edge nodes to pre-render high-frequency perspectives. Combined with the deferred rendering algorithm, 3D model rendering is processed in three stages to ensure a frame rate of ≥30fps.

[0061] In the above steps, the LOD adaptation standard is as follows: high-performance devices (GPU parameters ≥ 1024, CPU single-core score ≥ 1800) load a 1.2 million-face model with 8K textures; medium-performance devices (GPU ≥ 512, CPU ≥ 1200) load a 600,000-face model with 4K textures; and low-performance devices (GPU ≥ 256, CPU ≥ 800) load a 250,000-face model with 2K textures.

[0062] Personalized AR interaction adjustment: Collect the pupil diameter di, current ambient light intensity li, and light source color temperature Ti for the i-th visitor, and calculate the comprehensive display evaluation index αi using a weighted summation algorithm. The formula is as follows:

[0063]

[0064] Where d′, l′, and T′ represent the standard pupil diameter, standard illumination intensity, and standard color temperature, respectively, and κ1=0.4, κ2=0.3, and κ3=0.3, the AR image parameters are dynamically adjusted based on αi; the speech rate vi, sound decibel Bri, and environmental noise decibel Bit at the t-th acquisition time point are collected for the i-th visitor, and the speech adjustment evaluation coefficient βit is calculated using a weighted summation algorithm, the formula of which is:

[0065]

[0066] Where v′, Br′, and B′ represent standard speech rate, standard sound decibel, and standard noise decibel, respectively, and λ1=0.45, λ2=0.3, and λ3=0.25, the speech rate and volume are dynamically adjusted based on βit. When the user browses for more than 4 seconds, the ORB feature detection algorithm is used to match the AR image and text description and display them in a floating manner.

[0067] In the above steps, the standard parameter values ​​are:

[0068]

[0069] The evaluation index is shown. Calculation time ≤ 30 milliseconds, screen parameter adjustment response ≤ 100 milliseconds, voice adjustment evaluation coefficient. Calculation time ≤ 50 milliseconds, voice parameter adjustment delay ≤ 200 milliseconds;

[0070] The ORB feature detection algorithm extracts ≥500 feature points / frame, Hamming distance matching threshold ≤50, comparison time ≤150 milliseconds, text floating label size 300×150 pixels, automatically avoids the main body of the product (avoidance rate ≥95%), and click response ≤100 milliseconds.

[0071] Cache and system collaboration: The LRU cache eviction algorithm (where the cache capacity is 5% of the device storage) is used to evict idle blocks, and then the module interaction is realized through the gRPC protocol. An adaptive anomaly adjustment algorithm is used to handle loading timeouts, insufficient computing power, and network interruptions. The InfluxDB time-series data query algorithm is used to store and query system operation indicators.

[0072] In the above steps, the LRU cache eviction rate is ≥100MB / s, the total storage usage for browsing multiple exhibits is ≤500MB, and the adaptive anomaly adjustment strategy is to switch to low-precision backup content (core area 2K texture, 200,000-face model) when loading timeout (>3 seconds), reduce the number of facets of the model by 10% when the frame rate is <25fps, and enable local core area cache when the network is interrupted.

[0073] like Figure 2 As shown, the virtual-real integrated interactive tour system for digital museums according to Embodiment 1 of this application includes:

[0074] High-precision AR content (with textures of ≥4K resolution and models of ≥250,000 faces) is divided and encoded according to cultural background, and then subjected to two-dimensional block sorting, SHA-256 hash calculation and PBFT consensus on the Hyperledger Fabric 2.4 framework for on-chain processing.

[0075] User prediction and preloading module: Collects pedestrian flow and AR device data, runs a two-factor prediction model and a vector angle algorithm (using formulas). (Calculate the deviation angle), trigger preloading and execute multi-threaded loading.

[0076] Computing power adaptation rendering module: classifies device computing power levels, performs LOD adaptation, lightweight Diffusion pre-rendering and deferred rendering;

[0077] The computing power adaptation rendering module includes a computing power detection unit (dividing levels based on OpenGL ES and Geekbench parameters), a pre-rendering unit (deploying a Stable Diffusion XL Tiny fine-tuning model with an inference speed ≥10 frames / second), and a deferred rendering unit (Draw Call ≤500 / frame, texture sampling rate ≥2 times / pixel).

[0078] The interactive personalization adjustment module collects data from the i-th visitor. respectively through formula , calculate

[0079] Then, based on the calculation results, the display / voice parameters are adjusted, ORB detection is performed, and floating text descriptions are displayed.

[0080] The caching and coordination module performs LRU cache eviction, enables module interaction via gRPC, and runs adaptive exception handling and time-series data query algorithms.

[0081] The caching and collaboration module includes a data monitoring unit (with InfluxDB query response time ≤ 1 second).

[0082] Example 2 is illustrated with a specific implementation case:

[0083] Hardware: Visitor equipment includes high-performance HUAWEI VR Glass 2, medium-performance Xiaomi 14, and low-performance Redmi 12. Exhibition area equipment includes an edge computing node every 30 meters, a UWB positioning base station every 10 meters, BLE beacons embedded in the display cases, BH1750 light sensors, TCS34725 color temperature sensors, and MP34DT05 voice sensors. Consortium blockchain nodes include the museum's main node (deployed in the data center), two exhibition area edge nodes, and one technology service provider verification node.

[0084] Specific implementation: Taking the "Tricolor Camel Figurine" (assigned code Tang Dynasty-Silk Road-001) and "Maidservant" (assigned code Tang Dynasty-Court Life-002) as examples, the visual areas are divided according to the core area (main body of the cultural relic), the secondary core area (decorative details), and the edge area (background environment);

[0085] The 8K texture of the "Maidservant" painting was split into 24 blocks of 512×512 pixels (using ETC2 format, 50MB / block), the 1.2 million-face 3D model of the "Tricolor Camel Figurine" was split into 10 blocks of 120,000 faces each (using glTF2.0 format, 20MB / block), and the interactive information layer (cultural background, craft explanation) was split into 5 JSON blocks (6MB / block).

[0086] Deploy a Hyperledger Fabric 2.4 consortium blockchain, configure PBFT consensus parameters such as ≥2 / 3 node confirmation and block generation time of 1.5 seconds, generate a hash value for each block using the SHA-256 algorithm, and package it with the encoding and timestamp onto the blockchain to form a security verification library;

[0087] A lightweight Diffusion pre-rendering model, a finely tuned version of Stable DiffusionXL Tiny, with 1.2B parameters, was deployed at three edge nodes in the exhibition area. The training dataset contained 100,000 perspective images of Tang Dynasty artifacts plus 50,000 cultural annotations. A two-factor prediction model was deployed and trained using nearly three months of visitor flow data from the exhibition area, ensuring an accuracy of 88%.

[0088] UWB positioning and gyroscope data were calibrated, and a prediction threshold was set. The average movement time from the adjacent exhibition area (Tang Dynasty pottery area) to the target exhibit was 7 seconds. At the same time, the prediction threshold was 7 × 1.8 = 12.6 seconds. The vector angle algorithm error was calibrated to ±4° to ensure accurate determination of the line of sight direction. Then, multi-threaded loading was tested. Using Redmi Note12 with 4 loading threads, the core area loading time was 0.8 seconds and the full content loading time was 4.5 seconds, which met the solution indicators.

[0089] The devices were divided into different tiers based on their computing power. At the same time, LOD adaptation parameters were adjusted. The HUAWEI VR Glass2 was used to load a 1.2 million polygon model and 8K texture, the Xiaomi 14 was used to load a 600,000 polygon model and 4K texture, and the Redmi 12 was used to load a 250,000 polygon model and 2K texture, ensuring that the frame rate of all three devices remained stable at over 30fps.

[0090] High-frequency viewpoints were pre-rendered, and historical data was used to determine that the "0° front view and 30° right view (showing camel hump patterns)" of the "Tricolor Camel Figurine" and the "close-up of the face and clothing details" of the "Maid Figure" were high-frequency interactive content. After pre-rendering, the images were cached to edge nodes with a 70-millisecond delay. Then, standard parameters were set: d'=4.5mm (pupil diameter), l'=500lux (light), T'=5500K (color temperature), v'=120 words / minute (speech rate), Br'=70dB (volume), and B'=50dB (noise). The sensor acquisition accuracy was calibrated (pupil measurement error ±0.08mm, lighting accuracy ±15%).

[0091] We invited 200 tourists to test and collect feedback to optimize the rendering smoothness of low-computing-power devices (Redmi 12 frame rate increased from 30fps to 32fps), adjust the voice adjustment coefficient weight (λ3=0.25→0.3, enhancing noise adaptability), and optimize the LRU cache eviction policy (eviction speed increased to 120MB / s).

[0092] Among the visitors, one using a Xiaomi 14 spent 15 seconds (>12.6 seconds) in the Tang Dynasty pottery area, with their gaze directed towards the "Tricolor Camel Figurine" (vector angle θ=22°). The prediction model output probability was 0.85, and the edge node initiated preloading. When the visitor reached the exhibit, the core area had already been loaded, allowing for zooming to view the details of the camel hump patterns. During the loading process, the system verified that the hash value of the camel hump pattern block matched the on-chain record, ensuring that the content had not been tampered with.

[0093] Visitors using HUAWEI VR Glass 2 to view the 8K texture of "The Maidservant" exhibit experienced a deferred rendering algorithm with three stages of processing, a draw call rate of 450 frames, a texture sampling rate of 2.3 times per pixel, and a pre-rendered result when switching to the left 45° viewpoint, resulting in no ghosting.

[0094] Visitors using Redmi 12 loaded a 250,000-facet "Tricolor Camel Figurine" model with 2K textures at a frame rate of 32fps and a CPU usage of 28%. There was no lag during the browsing process, and the difference in experience compared to AR glasses was less than 180ms.

[0095] Visitors wearing AR glasses; exhibition area illumination l_i=1200 lux (direct sunlight); color temperature T_i=6200K; comprehensive display evaluation index calculated according to formula:

[0096] The system automatically adjusted the screen brightness by 15% and increased the contrast by 8%. Visitors reported that the murals had realistic colors and no glaring light. Meanwhile, the noise level in the exhibition area was 65dB, and the visitors spoke at a rate of 95 words per minute. The system calculated a voice adjustment coefficient of 26.3 and increased the speaking speed by 20% (144 words per minute) and the volume to 78dB, so that the explanations could still be heard clearly in the noisy environment.

[0097] Based on the feedback from the above-mentioned visitors, the average loading time of the core area of ​​this system is 0.9 seconds, the average loading time of the full content is 4.2 seconds, the interaction latency is ≤90 milliseconds, the experience difference between multiple devices is ≤200 milliseconds, the device adaptation rate is 100%, the content tampering recognition rate is 100%, the block transmission success rate is 99.9% (98.5% in weak network environment), and the visitor satisfaction survey shows that 89% of visitors think the details are clear and there is no lag, 82% of visitors think the voice explanation is clear and the visuals are comfortable, and the interactive participation rate is 65% higher than that of traditional exhibitions (see the table below).

[0098] Comparison Dimensions Traditional exhibition situation Effectiveness of the Implementation of this Technical Solution Improve / Optimize performance Core area loading time No high-resolution content; regular content takes ≥3 seconds to load. Average 0.9 seconds (≤1 second) Core area loading speed increased by 70% Full content loading time No high-resolution content, large file loading takes ≥10 minutes Average 4.2 seconds (≤5 seconds) Full load efficiency improved by over 99%. Interaction delay (zooming / view switching) Regular content interaction latency ≥300 milliseconds ≤90 milliseconds Interaction response speed improved by 70% Differences in experience across multiple devices Supports only high-end devices; low-performance devices may experience lag (difference ≥ 1 second). ≤200 milliseconds Improved consistency of experience across multiple devices by 80% Equipment compatibility Only supports flagship phones / professional devices (compatibility rate ≤60%) 100% (covering AR glasses, mid-range devices, and budget devices) Equipment compatibility expanded by 67%. Content tampering detection rate No security verification mechanism (recognition rate 0%) 100% Achieving zero risk in the security of highly refined cultural content Chunk transmission success rate No chunked transfer; large file transfer success rate ≤85%. 99.9% in normal environments and 98.5% in weak network environments. Transmission stability improved by 17%-19% Visitor satisfaction (clear details, no buffering) ≤50% 89% Satisfaction increased by 78% Visitor satisfaction (clear audio and comfortable visuals) ≤45% 82% Satisfaction increased by 82%. Interactive engagement Standard explanation / simple QR code scanning, participation rate ≤30% ≥95% (Active zoom, perspective switching, text viewing) Interactive engagement increased by 65%.

[0099] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0100] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0101] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A virtual-real integrated interactive tour method for digital museums, characterized by: Includes the following steps: (S1) Content Preprocessing and Security Management: Cultural backgrounds are divided according to the origin period of exhibits and historical events, and unique codes are assigned. A two-dimensional block sorting algorithm based on cultural relevance and visual weight is used to determine the block priority. AR content is split into texture layer, 3D model layer, and interactive information layer. Then, a unique hash value for each block is generated using the SHA-256 hash algorithm. The block ID and cultural background code are combined and packaged into a security verification package. A consortium blockchain is built using the Hyperledger Fabric 2.4 framework, and the PBFT consensus algorithm is used to put the verification package on the blockchain to ensure the integrity of the content. (S2) User behavior prediction and preloading: Integrate the traffic data of the exhibition area with the positioning, gyroscope and gaze data of AR devices, construct a two-factor prediction model through logistic regression, and combine the vector angle algorithm to determine the deviation angle. When the prediction probability is ≥0.7 or the dwell time is ≥prediction threshold, preloading is triggered. Multi-threaded concurrent loading is adopted. When the user deviates, the non-core block loading is interrupted and resources are released. (S3) Computing power adaptation and rendering: The computing power level of the device is divided by OpenGL ES parameters and Geekbench score. The LOD dynamic adaptation algorithm is used to match the corresponding precision 3D model and texture. The lightweight Diffusion pre-rendering algorithm is deployed at the edge node to pre-render high-frequency viewpoints. Combined with the deferred rendering algorithm, the 3D model rendering is processed in three stages to ensure the frame rate is ≥30fps. (S4) Personalized Adjustment of AR Interaction: Collect the pupil diameter di, current ambient light intensity li, and light source color temperature Ti of the i-th visitor, and calculate the comprehensive display evaluation index αi using a weighted summation algorithm. The formula is as follows: Where d′, l′, and T′ represent the standard pupil diameter, standard illumination intensity, and standard color temperature, respectively, and κ1=0.4, κ2=0.3, and κ3=0.3, the AR image parameters are dynamically adjusted based on αi; the speech rate vi, sound decibel Bri, and environmental noise decibel Bit at the t-th acquisition time point are collected for the i-th visitor, and the speech adjustment evaluation coefficient βit is calculated using a weighted summation algorithm, the formula of which is: Where v′, Br′, and B′ represent standard speech rate, standard sound decibel, and standard noise decibel, respectively, and λ1=0.45, λ2=0.3, and λ3=0.25, the speech rate and volume are dynamically adjusted based on βit. When the user browses for more than 4 seconds, the ORB feature detection algorithm is used to match the AR image and text description and display them in a floating manner. (S5) Cache and system collaboration: The LRU cache eviction algorithm is used to evict idle blocks, while the gRPC protocol is used to realize module interaction. An adaptive exception adjustment algorithm is used to handle loading timeouts, insufficient computing power, and network interruptions. The InfluxDB time series data query algorithm is used to store and query system operation indicators.

2. The method for interactive virtual and real-world tours of digital museums according to claim 1, characterized in that, In step (S1), the texture layer is divided into 24 512×512 pixel blocks at 8K resolution, the 3D model layer is divided into 10 blocks of 200,000-300,000 faces, and the interactive information layer is divided into 5 JSON format blocks.

3. The method for interactive virtual and real-world tours of digital museums according to claim 1, characterized in that, In step (S1), the consortium blockchain includes one museum master node, one edge node, and one technical service provider verification node. The PBFT consensus requires ≥2 / 3 of the nodes to reach an agreement, the block generation time is ≤2 seconds, and if the local hash verification is inconsistent during block loading, it will be retransmitted via HTTPS.

4. The method for interactive virtual and real-world tours of digital museums according to claim 1, characterized in that, In step (S2), the training data for the two-factor prediction model consists of pedestrian flow data and AR device data from the past three months. The model accuracy is ≥85%, and the calculation time for the vector angle algorithm is ≤30 milliseconds.

5. The method for interactive virtual and real-world tours of digital museums according to claim 1, characterized in that, In step (S3), the LOD adaptation standard is as follows: high computing power devices load a 1.2 million-face model and 8K texture; medium computing power devices load a 600,000-face model and 4K texture; and low computing power devices load a 250,000-face model and 2K texture.

6. The method for interactive virtual and real-world tours of digital museums according to claim 1, characterized in that, The standard parameter values ​​in step (S4) are: The evaluation index is shown. Calculation time ≤ 30 milliseconds, screen parameter adjustment response ≤ 100 milliseconds, voice adjustment evaluation coefficient. Calculation time ≤ 50 milliseconds, voice parameter adjustment delay ≤ 200 milliseconds.

7. The method for interactive virtual and real-world tours of digital museums according to claim 1, characterized in that, In step (S4), the ORB feature detection algorithm extracts ≥500 feature points / frame, the Hamming distance matching threshold is ≤50, the comparison time is ≤150 milliseconds, the size of the text floating label is 300×150 pixels, it automatically avoids the main body of the product, and the click response is ≤100 milliseconds.

8. The method for interactive virtual and real-world tours of digital museums according to claim 1, characterized in that, In step (S5), the LRU cache eviction rate is ≥100MB / s, the total storage usage for browsing multiple exhibits is ≤500MB, the adaptive anomaly adjustment strategy is to switch to low-precision backup content when loading timeout, and to gradually reduce the number of model faces by 10% when the frame rate is <25fps, and to enable the local core area cache when the network is interrupted.

9. The virtual-real integrated interactive tour system for digital museums according to any one of claims 1-8, characterized in that, include: The content preprocessing and security module divides and encodes high-precision AR content based on cultural background, and then performs two-dimensional block sorting, SHA-256 hash calculation, and PBFT consensus on the Hyperledger Fabric 2.4 framework for on-chain processing. The user prediction and preloading module collects data on pedestrian flow and AR devices, and runs a two-factor prediction model and a vector angle algorithm. The formula for the vector angle algorithm is: This triggers preloading and executes multi-threaded loading; The computing power adaptation rendering module divides the device computing power levels and performs LOD adaptation, lightweight Diffusion pre-rendering and deferred rendering. The interactive personalization adjustment module collects data from the i-th visitor. respectively through formula , calculate Then, based on the calculation results, the display / voice parameters are adjusted, ORB detection is performed, and floating text descriptions are displayed. The caching and coordination module performs LRU cache eviction, enables module interaction via gRPC, and runs adaptive exception handling and time-series data query algorithms.

10. The virtual-real integrated interactive tour system for digital museums according to claim 9, characterized in that, The computing power adaptation rendering module includes a computing power detection unit, a pre-rendering unit, and a deferred rendering unit, while the caching and collaboration module includes a data monitoring unit.

Images