DESIGNING AN AUTOMATED PHOTOGRAMMETRY PIPELINE BY HIGHER EDUCATION STUDENTS CONSIDERING HARDWARE COMPONENTS BASED ON CLOUD LINUX SERVERS

Abstract

The article examines the problem of digitalizing cultural and historical heritage sites in the Chernihiv region, which are currently under threat due to armed aggression and natural factors. The research is based on the activities of the student research club «Artificial Intelligence: from Theory to Startup» and the Innovation Hub of the T.H. Shevchenko National University «Chernihiv Colehium». Primary attention is paid to photogrammetry technology, which requires significant computing power to create high-precision threedimensional copies of real objects. The work describes in detail the transition from sequential computing on the CPU to massively parallel computing on the GPU using CUDA cores to accelerate Structure from Motion algorithms and dense 3D reconstruction. Hardware requirements for video memory volumes and PCIe Gen4/5 bus bandwidth for processing large arrays of raster data are considered. The process of building automated pipelines through the system administration of Linux servers is highlighted, which allows for operation in a mode without a graphical interface, thereby freeing up critical hardware resources for computations. An important component is the description of the integration of deep learning models for semantic segmentation and clearing images of digital noise and artifacts. Aim of the work is the theoretical justification and design of the architecture of an automated photogrammetry pipeline in a cloud environment based on an operating system with a Linux kernel, using artificial intelligence technologies and containerization to optimize the process of creating digital copies of cultural heritage objects and improving the training of IT specialists.

Methodology. The research methodology includes a search, systematic theoretical analysis, and systematization of scientific works and practical experience regarding the use of photogrammetry and cloud computing. To solve technical problems, SfM/MVS computer modeling methods, Docker containerization technology for isolating software environments using the NVIDIA Container Toolkit, and interpretation of the results of the pipeline testing on the example of objects from the National Architectural and Historical Reserve «Chernihiv Ancient» were applied. In addition to the technical analysis, a method for developing methodological recommendations for implementing this technology stack into the educational process of training students in the specialty «Professional Education (Digital Technologies)» was used.

Scientific novelty of the obtained results lies in the improvement of a comprehensive model of an automated photogrammetric pipeline that combines cloud Linux servers, Docker isolation of CUDA environments, and intelligent preprocessing of data by neural networks into a single work cycle. This allows for the elimination of software dependency version conflicts and ensures process stability when using various software such as Meshroom, Blender, and PyTorch. The method of creating digital twins of historical heritage objects has been improved by automating the stages of clearing the raster from digital noise using Tensor cores, which ensures higher metric accuracy of models with a significant reduction in time costs. The methodology for creating interdisciplinary connections during the implementation of students' creative projects has been further developed, integrating knowledge of computer architecture, operating systems, and graphic design.

Conclusions. The study confirms that the use of Linux-based operating systems proves its advantage due to efficient GPU resource management and the ability to work via the command line, which frees up to 10-15% of video memory for photogrammetry needs. The implementation of the Docker Engine together with the NVIDIA Container Toolkit ensures the stability of the computing pipeline and the possibility of simultaneous use of different CUDA versions. The implementation of such a project is a powerful educational tool that prepares students for real challenges in the fields of DevOps for game development, digital archiving, and computational design, shifting the emphasis from routine processing to the design of complex automated systems. The practical application of the developed model allows for the creation of reliable digital archives of Chernihiv monuments, which can be used for future restoration, 3D printing, or the creation of interactive augmented reality applications. A promising direction for further research is the automation of model conversion into formats optimized for mobile devices and VR platforms.