ANALYSIS OF COMPUTATIONAL INTELLIGENCE METHODS FOR MODELING, IDENTIFICATION, OPTIMIZATION OF SYSTEMS AND DECISION SUPPORT

Authors

DOI:

https://doi.org/10.20998/2413-3000.2024.9.5

Keywords:

remote identification of dynamic objects, object detection, optical flow, speed identification, deep learning, convolutional neural networks, processing, video, decision support system, computational intelligence

Abstract

The latest methods and tools of computational intelligence that have found widespread application in various fields, including information control systems, decision support systems, as well as modeling and remote identification of dynamic system, have been analyzed. Special attention is given to methods such as HunyuanVideo, emg2pose, StableAnimator, DEYO, YOLOv11, YOLO-NAS, SynCamMaster, FlowNet, Momentum-GS, Liger-Kernel, Stereo Anywhere, and Neural Attention Memory Models. The analysis shows the great potential of these technologies for improving existing solutions in the field of computational intelligence. HunyuanVideo uses diffusion models for video generation, significantly improving visualization and dynamics while reducing computational power requirements. The emg2pose and StableAnimator methods provide high precision and flexibility, which are especially important for real-time decision support systems. The application of technologies such as DEYO and YOLOv11 has improved the speed and accuracy of object detection, which is crucial for security and real-time video stream monitoring. The FlowNet and FlowNet 2.0 methods for optical flow estimation allow precise tracking of object motion, significantly improving the accuracy in dynamic scene processing. SynCamMaster synchronizes video from different viewpoints, opening up new opportunities for 3D scene reconstruction, demonstrated through the use of technologies like Momentum-GS. At the same time, specialized strategies such as Liger-Kernel are actively applied to enhance efficiency in complex environments like autonomous vehicles and robotics. The necessity of optimizing computational processes for integrating these methods into real-world systems is discussed, with a focus on ensuring high precision and speed of technology operation under resource constraints. The use of these technologies will enable the creation of innovative approaches to solving complex real-time problems, significantly improving the effectiveness and accuracy of existing systems. The included tables demonstrate the importance of integrating new technologies into various research fields.

References

Krzywanski J., Sosnowski M., Grabowska K., Zylka A., Lasek L., Kijo-Kleczkowska A. Advanced Computational Methods for Modeling, Prediction and Optimization—A Review. Materials, 2024. 17(14). 3521. URL: https://doi.org/10.3390/ma17143521 (дата звертання: 13.12.2024)

Нікуліна О. М., Северин В. П., Кондратов О. М. , Рекова Н. Ю. Аналіз інформаційних технологій для дистанційної ідентифікації динамічних об’єктів. Вісник НТУ «ХПІ». Серія: Системний аналіз, управління та інформаційні технології. Харків. НТУ «ХПІ», 2023. № 1 (9). С. 110–115.

HunyuanVideo: A Systematic Framework For Large Video. URL: https://doi.org/10.48550/arXiv.2412.03603 (дата звертання: 13.12.2024)

emg2pose: A Large and Diverse Benchmark for Surface Electromyographic Hand Pose Estimation. URL: https://doi.org/10.48550/arXiv.2412.02725 (дата звертання: 13.12.2024)

StableAnimator: High-Quality Identity-Preserving Human Image Animation. URL: https://doi.org/10.48550/arXiv.2411.17697 (дата звертання: 13.12.2024)

DEYO: DETR with YOLO for End-to-End Object Detection. URL: https://doi.org/10.48550/arXiv.2402.16370 (дата звертання: 13.12.2024)

YOLOv11: An Overview of the Key Architectural Enhancements. URL: https://doi.org/10.48550/arXiv.2410.17725 (дата звертання: 13.12.2024)

A Comprehensive Review of YOLO Architectures in Computer Vision: From YOLOv1 to YOLOv8 and YOLO-NAS. URL: https://doi.org/10.3390/make5040083 (дата звертання: 13.12.2024)

SynCamMaster: Synchronizing Multi-Camera Video Generation from Diverse Viewpoints. URL: https://doi.org/10.48550/arXiv.2412.07760 (дата звертання: 13.12.2024)

FlowNet: Learning Optical Flow with Convolutional Networks. URL: https://doi.org/10.48550/arXiv.1504.06852 (дата звертання: 13.12.2024)

FlowNet 2.0: Evolution of Optical Flow Estimation with Deep Networks. URL: https://doi.org/10.48550/arXiv.1612.01925 (дата звертання: 13.12.2024)

Momentum-GS: Momentum Gaussian Self-Distillation for High-Quality Large Scene Reconstruction. URL: https://doi.org/10.48550/arXiv.2412.04887 (дата звертання: 13.12.2024)

Liger Kernel: Efficient Triton Kernels for LLM Training. URL: https://doi.org/10.48550/arXiv.2410.10989 (дата звертання: 13.12.2024)

Stereo Anywhere: Robust Zero-Shot Deep Stereo Matching Even Where Either Stereo or Mono Fail. URL: https://doi.org/10.48550/arXiv.2412.04472 (дата звертання: 13.12.2024)

An Evolved Universal Transformer Memory. URL: https://doi.org/10.48550/arXiv.2410.13166 (дата звертання: 13.12.2024)

Нікуліна О. М., Кондратов О. М. Методи дистанційної ідентифікації динамічних параметрів об’єкта. Інформаційні технології: наука, техніка, технологія, освіта, здоров’я: Тези доповідей ХXХІ міжнародної науково-практичної конференції MicroCAD-2023, 17-20 травня 2023 р. Харків, НТУ «ХПІ». 2023. С. 1047.

Published

2025-03-17