3D Reconstruction of Ancient Building Structure Scene Based on Computer Image Recognition

With the extensive application of computer image recognition (CIR), the high cost of three-dimensional (3D) models, long construction cycles, poor data visualization, and other problems have become the main bottlenecks in further development of CIR. Artificial intelligence (AI) is an important branch of computer science and has a wide range of application prospects and high practical value, especially in the field of medical and health applications of intelligent machines. This article introduces the background of 3D reconstruction of ancient architectural structure scenes, and then presents academic research and a summary on two key applications of CIR. It then summarizes 3D reconstruction and media technology in combination with AI used for medical diagnoses. In this article, the algorithm model is established, and various algorithms are proposed to provide a theoretical basis for the research of 3D reconstruction of ancient building structure scenes based on CIR.

reconstruction configuration of room size , with some finding 3D modeling of the indoor environment playing an important role in various applications such as indoor navigation, building information modeling, and interactive visualization (Cui, 2019). The framework also provides an end-to-end automatic processing method for mapping the initial point cloud to the specified classification results (Zhang & Liang, 2017). The precision 3D measurement system has been thriving because of the research conducted in recent years. Most laser scanners are based on laser scanning technology that can directly obtain 3D data in real time (Sung & Lin, 2017). Depth learning has also been used to reconstruct the 3D model of structural perceptual semantics of cable-stayed bridges. The traditional method of reconstructing the bridge semantic 3D model is usually unable to reconstruct the structure-aware semantic 3D model when using low-quality point clouds (Hu, 2021). The studies have achieved satisfactory results but nonetheless present some problems despite continuous technology updates.
The application of CIR in 3D reconstruction of building structure scenes has been analyzed at different levels by many scholars. Research has systematically discussed and studied the integrity and efficiency of urban outdoor building scene reconstruction (Rebecq, 2018). Damage identification has been conducted in complex scenes and structural surfaces by combining 3D reconstruction technology and digital processing (Fan, 2019). Automatic and semi-automatic reconstruction technology has also been studied based on multiple images by applying the principle of multiple view geometry in computer vision, subsequently applying clustering in pattern recognition and correlation in text retrieval to reconstruction. This process includes fast similarity calculation, visualization, and semi-automatic reconstruction (Bittner, 2018). The recent report on building renovation also posits that there is still work to be done to fully address the issue of sustainability. From the architectural transformation perspective, goals of comprehensive sustainability thus become salient points for discussion (Kamari et al., 2017). A single image stereo reconstruction algorithm based on structured scenes for buildings that cannot be reconstructed by lasers or multiple images has been re-created by existing studies (Han et al., 2019). Applications for military, civilian, and other sectors have been explored, which shows that demand for 3D reconstruction is increasing, and the research on digital models is an important application (Di Ludovico, 2017). These studies show that applying CIR has a positive effect; although, some problems still remain.
3D scene reconstruction of buildings has gradually risen to a higher level with the development of modern science and technology. Following CIR technology, this study discusses an ancient architectural structure based on 3D representation. Analysis of the relevant factors and influencing factors of CIR was combined with the algorithm model and subsequently simulated. CIR was found to be very suitable, and therefore, has a certain development prospect.

Introduction of Image Recognition Technology
Image recognition is a new AI technology and has three stages of development: text recognition, image processing and recognition, and object recognition (Miller & Eric, 2018). Image recognition is the processing and analysis of images to determine the objects to be investigated. Image recognition is a human function and can also be realized through computer technology. Although a person's recognition ability is strong, human cognitive ability cannot meet the demand in increasingly complex scenarios, which has led to the development of CIR technology. This is akin to studying human cells; it is impossible to see individual cells without the use of precision instruments such as microscopes. When there are some requirements that cannot be met by technologies in a certain field, innovative technologies emerge. Image recognition technology can process massive amounts of physical information and solve problems that are not apparent to humans.

Principle of Image Recognition Technology
Image recognition technology is simple in principle; the only complexity is if the information it must process is too bloated (Guo & Bin, 2018;Krittanawong, 2017). Computer processing technology is inspired by real life and simulated by software. This emphasizes that no significant difference, in principle, exists between CIR technology and human image recognition. The difference is that machines do not have human senses. Human visual recognition is not simply based on memories stored in the brain-images are classified according to the characteristics of the pictures themselves and are then classified according to type. The principle of image recognition technology is shown in Figure 1.
When a human sees a picture, the brain immediately responds to it and judges whether it has ever seen something similar (Wagner, 2019). This is known as fast identification between perception and vision, and the subsequent process of identification is similar to a search. The same is true for machine image recognition technology, which classifies important features and eliminates redundant information to identify an image. These characteristics extracted by the machine are sometimes significant and common, thereby greatly affecting the recognition speed of the computer. Therefore, in computer vision recognition, the content of an image is represented by its characteristics.

Pattern Recognition
Pattern recognition is a key technology in the field of computer technology. Among other applications, it describes, distinguishes, and classifies by analyzing and processing various forms of information (He, 2019).
CIR technology is used for human body image recognition. Pattern recognition is a type of basic intelligence necessary in image recognition. The development of more complex computer technology and the emergence of AI, however, has outpaced the ability of the human brain. So, people expect to either replace or expand human mental work with computers (Rong, 2020). Thus, computer pattern recognition was developed. Essentially, pattern recognition classifies data mathematically using probability and statistics. Pattern recognition includes statistical and syntactic pattern recognition.

MAIN CHARACTeRISTICS oF CoMPUTeR INTeLLIGeNT IMAGe ReCoGNITIoN TeCHNoLoGy
The main features of computer intelligent image recognition technology are shown in Figure 2.

Large Amount of Information Storage
CIR technology is widely used in various fields because of its capacity to store an abundance of information. Because the system structure and memory specification are similar, the matching image can be selected according to distinct characteristics during image recognition without affecting the recognition speed, therein ensuring the effectiveness of recognition.

Relatively Strong Correlation
All stored images can be intelligently processed through computer images, speeding up image processing. All images can also be compressed and classified, which ensures the stability of the system and reduces the error in the recognition process. This enables the system to better play a scientific and effective management function. The method effectively solves the clutter problem and other related problems in image recognition.

Highly Artificial
Programmers can accurately define the role of the computer when writing computer programs. Once the computer starts, programmers must manipulate it, thereby creating the human factor in the computer. After the computer's intelligent image recognition system starts, the image can be detected manually. However, the incomplete image cannot be recognized, thus leading to careless processing, which affects the accuracy of the entire process. After the computer has completed the identification work, its identification results must be evaluated manually. Errors in human judgment also exist, causing artificial misjudgment. This phenomenon has affected people's daily lives; therefore, measures are needed to ensure the stability and reliability of the computer's intelligent mode.

Concept of 3D Reconstruction
Three-dimensional reconstruction is used to establish mathematical models of 3D objects that conform to computer expression and processing and to process, operate, and analyze them. It is an important means to realize virtual reality and an expression of the objective world.

Figure 2. Main characteristics of computer intelligent image recognition technology
3D reconstruction constitutes 3D information from one or multiple perspectives of computer vision. Because the information obtained from a single image is incomplete, empirical rules must be prepared for 3D reconstruction. In contrast, multiple view 3D reconstruction is relatively easy; it analyzes the relationship between the measuring camera coordinate system and the coordinate system by first correcting the camera. Accordingly, 3D reconstruction of several two-dimensional (2D) images is thus carried out.

Theoretical overview of 3D Reconstruction of CIR
The 3D reconstruction process of 5-point CIR is summarized as shown in Figure 3.

Image Acquisition
Before image processing, 2D images of 3D objects are captured by cameras. The lighting conditions and geometric characteristics of the camera impact the image processing.

Camera Calibration
Camera calibration is used to create effective image models, solve internal and external camera parameters to obtain 3D point coordinates in space, and provide image matching results for 3D reconstruction.

Feature Extraction
Features include feature points, feature lines, and regions. In many cases, feature points are mapping primitives, and the extraction form of these feature points is closely related to the mapping strategy used. Therefore, when extracting feature points, the computer must first decide which matching strategy to use.

Stereo Matching
Stereo space detection compares two images according to the extracted features. The same physical space representation points of two images are then matched one by one. The interference caused by variables must be considered in the matching process including lighting conditions, noise, scene deformation, and physical surface and camera device characteristics.

3D Reconstruction
Combining the calibration parameters of internal and external cameras, more accurate matching results can be obtained to retrieve 3D scene information. This is because the accuracy of 3D reconstruction is affected by the accuracy of matching and the error of internal and external camera parameters. To realize the stereo vision system, it is necessary to design highly accurate and error free correspondence, related displacement, and geometric knowledge.

Application of AI and Multimedia Technology for Medical Diagnoses Based on Image Recognition
The application of AI and multimedia technology of image recognition for medical diagnoses is shown in Figure 4.

Fast Processing of Image Data
If there is an abundance computations required to process a large number of images simultaneously, people are bound to make mistakes due to overwork.

Rich Types
Various kinds of diseases, from cardiovascular diseases to tumors, need imaging and recognition. Human factors limit doctors' proficiency in recognizing multiple disorders. The high efficiency and huge amount of data a computer can process, however, enables it to recognize and process images of a multitude of diseases.

Combination of Image Recognition and Big Data
AI is not limited to patients' image data; it can also digitize patients' medical records, genes, family history, and other data. It is able to integrate diet habits, work and rest time, and other important information into the model, thereby making a more accurate and personalized diagnosis and prediction. The medical diagnosis robot incorporates these functions. It reads medical images and analyzes and provides a diagnosis through the electronic medical record database. Additionally, robots can also learn patients' living habits, CIR, and symptoms by communicating with them.

ALGoRITHM MoDeL FoR 3D ReCoNSTRUCTIoN Projection Model Calibration Algorithm
The point ( , , ) i j p T on the image plane with camera focal length f and principal point coordinates (1, 1) corresponds to a projection curve passing through the origin of the camera coordinate system, and the curve direction is ( , , ) i j p T . Different curve shapes represent different projection patterns, and ( , , ( , )) i j f i j T has different manifestations. For pinhole projection models, the following is used: If the lens distortion parameter of the unified projection model is set as 1, then The homogeneous coordinate of the feature point is f i j p x ( , ) = − 2 1 2 2 ρ . The relationship between this 3D point and the projection point on the image plane is, thus, as follows: The upper form is multiplied by the left and right sides by the lower form.
In the formula, v is a scale parameter and 0 The coordinate disk can be used to determine the corresponding relationship between the coordinates in the coordinate system and the 2D points of the image, allowing acquisition of the above parameters.

Structural Reconstruction Algorithm
The coordinate of any point on a straight line on the rigid body at the time point is i r x r y t v r x r y t ( ) (  )   21  22  2  11 12 1 0 + + − + + = , and its translational velocity is P x y z ( , , ) . Because vector m is perpendicular to the plane, the following vector relationship can be obtained: Taking the derivative of both sides of the above formula simultaneously and integrating formula 8, the following expression transpires: Substituting the perspective projection formula based on pixel points into the following formula allows for Collating it then leads to y fy z = / In the formula above, the (x, y) coordinate is the depth of the point, the motion speed of the rigid body is related to the Z coordinate value on the detected rigid body line, and its motion speed is independent of the angular velocity.

SIMULATIoN eXPeRIMeNT oF 3D ReCoNSTRUCTIoN oF ANCIeNT BUILDING STRUCTURe SCeNeS
Empirical research was used to study the simulation experiment of 3D reconstruction of an ancient building structure scene. The objective is to select four teams to study 3D reconstruction of ancient building structure scenes, with a data sample of 25 people from each team. Through image accuracy analysis, image integrity analysis, 3D reconstruction content analysis, and tool use analysis, the newer method was found to be more advantageous compared to the traditional. Table 1 shows the 3D points obtained by a 3D scanner.

Image Accuracy Analysis
The main contents of CIR in 3D reconstruction include image acquisition, camera calibration, feature extraction, stereo matching, and 3D reconstruction. Traditionally, contents are only related to maps, photos, and videos. Figure 5 shows the image accuracy analysis of new and traditional methods regarding ancient buildings. Figure 5 shows that the image accuracy of the 3D reconstruction of the structural scene of the four ancient buildings using the new method is higher compared to the traditional way. The image accuracy using the new method in ancient building A is 97.36%, and the image accuracy of the traditional method in ancient building A is 89.14%. Meanwhile, the image accuracy of the new method in ancient building B is 96.47%, and that of the traditional method in ancient building B is 87.24%. The image accuracy of the new method in ancient building C is higher at 97.13%, and the image accuracy of the traditional method in ancient building C is lower at 88.54%. The image accuracy of the new method in ancient building D is also higher at 96.12%, compared to the image accuracy of the traditional method in ancient building D at 87.64%. The image accuracy of the new method in the four ancient buildings is 96.77% on average, which is contrasts with the image accuracy of the traditional method in the four ancient buildings at 88.14%. Calculations show that the accuracy of the new method is 8.63% higher than that of the traditional method.

Image Integrity Analysis
This paper summarizes the overall structure of the image with the following lines below. From a composition perspective, it is a useful way to find the lines in the composition. This allows for the description of the visual elements related to the image through one or more lines, or the connection of various visual elements scattered in the image through abstract lines to form a complete visual structure. Figure 6 shows the image integrity analysis of four ancient buildings in two ways, with values ranging from 1 to 10. Figure 6 also shows the image integrity of the new mode is higher compared with the traditional mode. Among these comparisons, the image integrity of the new way of the four ancient buildings is 8, 8.5, 9.5 and 9 for buildings A, B, C, and D, respectively. Meanwhile, the image integrity of the traditional way of the four ancient buildings is 7, 6.5, 6 and 7.5 for buildings A, B, C, and D,

CoNTeNT ANALySIS oF 3D ReCoNSTRUCTIoN oF ANCIeNT ARCHITeCTURAL STRUCTURe SCeNeS
The 3D reconstruction of the ancient architectural structure scene includes four primary aspects. Three-dimensional data acquisition automatically collects, signals, and sends data to the computer for analysis and processing through analog and digital measuring devices such as sensors and other such devices. The establishment mode uses corresponding drawing software to assist the establishment. Directional calculation also uses the information obtained from 3D data to conduct targeted evaluation. Model visualization is a way of thinking that uses ideas about reality to organize patterns. Figure 7 shows the utilization rate analysis of traditional buildings and ancient buildings on 3D data acquisition, model building, directional calculation, and model visualization. Figure 7 shows that the utilization rate of traditional buildings in quantitative calculation is high, although the ancient buildings have a higher utilization rate in the categories of building models and model visualization. The data shows that the utilization rate of traditional buildings in 3D data acquisition is 18%, while that of ancient buildings in is 9%. The utilization rate of traditional buildings in building models is 11%, while that of ancient buildings is 53%. The utilization rate of traditional buildings in directional calculation is notably high at 65%, whereas that of ancient buildings is 10%. And the utilization rate of traditional buildings in model visualization is 6%, and that of ancient buildings is 28%.

Tool Utilization Analysis
Following the above experimental research on 3D reconstruction of ancient architectural structure scenes recognized by computer image, this section analyzes the number of users of the following 3D reconstruction tools: 3D Studio Max (3DMAX), Autodesk Maya (Maya), Autodesk Computer Aided Design (AutoCAD), and Unigraphics NX (UG) as shown in Figure 8. Figure 8 shows that the four curves of the broken line chart comprise the first curve and the trend is downward. The trend of the second curve declines first, then rises, and then rapidly declines. The trend of the third curve, meanwhile, rises quickly, then falls, and then rises gradually. The fourth curve falls, then rises gradually, and finally rises steeply. The data reflects 14 people in Team 1 using 3DMAX, four people using Maya, three people using AutoCAD, and four people using UG. In Team 2, seven people used 3DMAX, three used Maya, 13 used AutoCAD, and two used UG. In Team 3, four people used 3DMAX, 11 used Maya, five used AutoCAD, and five used UG. In Team 4, 3 people used 3DMAX, two used Maya, six used AutoCAD, and 14 used UG. 3DMAX notably has the highest number of users in Team 1, Maya has the highest number of users in Team 3, AutoCAD has the highest number of users in Team 2, and UG has the highest number of users in Team 4. Therefore, experimental results show that the accuracy of 3D reconstruction of ancient building structure scenes using CIR is higher compared to traditional methods.

CoNCLUSIoN
Reconstruction of outdoor scenes from images is an important research field in computer vision and has a wide range of applications in areas such as virtual reality, urban digital modeling, and cultural heritage protection. Due to inhibiting factors, however, it is difficult to completely reconstruct the scene structure using traditional reconstruction methods. To overcome these difficulties, this paper analyzes the 3D reconstruction of ancient building structures based on computer vision. The analysis of image accuracy, image integrity, 3D reconstruction content, and tool use show that the results are better compared with traditional reconstruction methods.