Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{205001,
        author = {Vedant Prafulla Pandey and Isha Chaudhari and Rutuja Kshirsagar and Rajkumar Rathod and Harshvardhan Kulkarni},
        title = {Next-Generation Visual Perception: A Systematic Review of Transformer Architectures, Gesture Analysis, and Agentic AI Pathways},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {1},
        pages = {5553-5559},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=205001},
        abstract = {The paradigm of computer vision (CV) has undergone substantial conceptual evolutions, migrating from deterministic rule-based pixel manipulation to self-assembling deep feature representations, and ultimately to global self-attention networks. This systematic survey delineates the trajectory of visual pattern recognition frameworks, focusing on the synthesis of gesture tracking, micro expression decoding, and the deployment of Vision Transformers (ViTs). Historically constrained by laborious feature engineering sensitive to environmental occlusions, visual systems achieved robust translation invariance through Convolutional Neural Networks (CNNs). However, the localized inductive bias of convolutions restricts the acquisition of long-range contextual relationships. The emergence of the post convolutional attention paradigm processes images as discretized patch sequences, utilizing Multiheaded Self Attention (MHSA) to map holistic context. We analyze these mathematical structures against benchmark data and survey the burgeoning domain of Agentic AI, where passive inference is replaced by interactive autonomous loops featuring tool integration, iterative visual refinement, and Large Language Model (LLM) orchestration. Additionally, we contrast sensory backbones spanning RGB, volumetric depth (RGBD), electromyography (EMG), and inertial frameworks, alongside multimodal alignment schemas. Our synthesis maps current computational, infrastructural, and regulatory limitations, establishing an analytical blueprint for future autonomous vision frameworks.},
        keywords = {Computer Vision, Gesture Analysis, Vision Transformers, Agentic Systems, Human Computer Interaction, Attentional Networks, Deep Learning},
        month = {June},
        }