AI in Computer Vision

Tanushree Ray-22053822, Imad Parvez-22053776, Poorvi Singh-22054149, Pratham Kumar-22052836

[email protected], [email protected], [email protected], [email protected]
School of Computer Engineering, Kalinga Institute of Industrial Technology (KIIT), Bhubaneswar, Odisha- 751024

Abstract - external servers. The authors place a strong emphasis on

practical uses such healthcare diagnostics, driverless
navigation, and intelligent surveillance. In order to boost
Index Terms - Artificial Intelligence, Computer Vision, computational efficiency, the study concludes by outlining
Deep Learning, Image Processing, Neural Networks future research paths that enable advancements in edge-
based AI architectures and hybrid models that smoothly mix
local and cloud inference.
INTRODUCTION
2. Artificial Intelligence and Computer Vision – Sunila
The integration of Artificial Intelligence (AI) in computer Gollapudi
vision has led to transformative changes, enhancing the
ability of systems to process and analyze visual data.
Computer vision involves enabling machines to interpret This book chapter offers a thorough overview of AI-driven
images, videos, and other visual inputs, mimicking human computer vision, going over fundamental ideas, real-world
vision capabilities. With the advent of AI, particularly deep uses, and recent developments in the field. It separates
learning techniques, computer vision has achieved artificial intelligence (AI) into theory of mind AI, reactive
unprecedented levels of accuracy and efficiency. systems, restricted memory models, and self-aware AI. It
also makes a distinction between narrow AI (NAI), artificial
general intelligence (AGI), and superintelligent AI (SAI).
This paper aims to provide an overview of AI-driven
The article describes how deep learning models, such
computer vision, highlighting significant advancements,
convolutional neural networks (CNNs), are used in modern
applications, and challenges.
computer vision for tasks like motion analysis, object
detection, segmentation, and image classification. In his
RELATED WORKS exploration of computer vision architectures, the author
emphasizes how OpenCV, TensorFlow, and PyTorch may
1. Accelerated Computer Vision Inference with AI on the speed up development.
Edge Medical imaging, industrial automation, face recognition,
and autonomous driving are among the most significant uses.
The necessity and advantages of artificial intelligence (AI) The impact of optical feature recognition (OCR) and
for real-time computer vision applications are covered in intelligent character recognition (ICR) on automation and
this study, along with problems including high processing documentation is also covered in this chapter. Examine
demands, network latency, and bandwidth limitations. Many issues with your computer's vision, including data scarcity,
edge devices, like drones, embedded systems, and Internet confrontation attacks, model explanations, and potential
of Things cameras, cannot maintain the continuous fixes including transmission and the creation of synthetic
connectivity needed by traditional cloud-based AI models. data. According to the article's conclusion, the industry will
OpenVINO, a toolbox created to optimize deep learning soon adopt AI-led visual perception systems as a result of
models for edge deployment, is highlighted in the article. the synergistic role of computer vision, robot technology,
Hardware-friendly inference is made possible by and natural language processing (NLP). The future
OpenVINO, which lowers latency and boosts processing computer that powers the artificial intelligence that will be
effectiveness. The systematic approach for deploying AI used in the internet and edge computing is also briefly
models at the edge is described in the study, including examined.
topics like hardware integration, inference acceleration, and
model optimization. It presents the main components of 3. When, Where, and Which? Navigating the Intersection of
OpenVINO, such as the deployment manager, inference Computer Vision and Generative AI for Strategic Business
engine, and model optimizer, which aid in transforming Integration
deep learning models into a more effective structure.
The report also covers the benefits of edge computing in This paper addresses business generation and computer
terms of privacy and security because data processing takes vision integration in business applications, with a focus on
place locally rather than needing to be transmitted to
the strategic application of these technologies across many to 2021 and forecasting future trends. AlexNet, ResNet,
industries. This is separated into narrow AI, general AI, and VGGNet, and EfficientNet are among the eight major
super AI, and their characteristics in terms of creativity, architectures highlighted, which have greatly enhanced CV
automation, and estimated analysis are explained. Important applications such image restoration, visual tracking,
developments in generative AI are highlighted in the study, semantic segmentation, and recognition. DL models are
including transformer-based models (e.g., GPT, Stable more accurate, efficient, and adaptable to the actual world,
Diffusion, DALL·E) and their uses in automated content and lightweight architectures like MobileNets make it
creation, picture enhancement, and synthetic data output. By possible for embedded and mobile applications.
pointing out that generative AI creates original material, but Additionally, the study discusses ethical issues like
computer vision extracts and interprets visual data, it sets deepfakes, bias, and data privacy. Scalable models, cross-
generative AI apart from typical computer vision. domain integration, and moral AI practices are the primary
The study offers practical applications in the fields of areas of future study to increase transparency and equity.
healthcare, retail, autonomous cars, and surveillance,
demonstrating how AI may boost customer engagement, 6. Computer Vision Algorithms and Hardware
streamline supply chains, and improve decision-making. Implementations: A Survey
The study promotes the appropriate use of AI and covers
ethical concerns like deepfake misuse, bias in AI models, This paper gives a comprehensive review of computer
and data privacy threats. Before deployment, a structured AI vision techniques and their hardware implementations. It
selection methodology is suggested to assist businesses in discusses deep learning-based methods for object
evaluating domain-specific limitations, computing needs, identification, image segmentation, and image classification.
and data preparedness. In the near future, automation and The paper emphasizes how convolutional neural networks
human-AI collaboration will be redefined by hybrid AI (CNNs) such as AlexNet, VGGNet, and ResNet have
systems that integrate generative AI with computer vision, advanced and greatly increased accuracy in tasks involving
according to the authors. visual perception.
In terms of hardware, the study looks at how specialized AI
4. Artificial Intelligence in Colorectal Cancer Surgery: accelerators (such TPUs), GPUs, and FPGAs can improve
Present and Future Perspectives computing efficiency. It talks about optimization strategies
to boost performance, like quantization, model compression,
With an emphasis on intraoperative care, perception of the and energy-efficient designs. More energy-efficient
surgical process, and AI-based decision-making, this article hardware designs, poorly supervised learning, and small
examines the use of artificial intelligence (AI) and computer deep learning models for real-time applications are the main
vision (CV) in colorectal cancer (CRC) surgery. AI has areas of focus for future research.
demonstrated significant promise in CRC screening and
staging, particularly in image-based lymph node analysis 7. Computer Vision and Image Processing: A Paper Review
and polyp detection. Its intraoperative use is still very new,
nevertheless.According to the findings, even in complex Reviewing current developments in computer vision and
procedures like transanal total, medial mesothelial excision image processing, the study divides the area into four
(TaTME), CV-based AI models can accurately identify categories: deep learning, machine learning, object
surgical stages and movements. In order to support real-time identification, and image processing. It draws attention to
surgical decision-making, AI-driven models have been fundamental methods including segmentation, edge
trained to identify the best dissection planes for total detection, and feature extraction that form the basis of
mesorectal excision (TME) and analyze fluorescence signals contemporary vision systems. The study compares methods
for perfusion assessment. The function of AI in automated based on computational cost, accuracy, and efficiency to
surgical skill assessment is also included in the paper. Deep investigate how CNNs and GANs might improve object
learning is used to assess procedural accuracy and motion identification and classification accuracy. Medical imaging,
efficiency. Lack of established annotation techniques, self-driving cars, security monitoring, and robotics are
inconsistent surgical process, and inadequate multi-center important uses. Occlusions, illumination fluctuations, and
validation are major obstacles. Larger annotated datasets, AI real-time processing limitations are among the other
models that incorporate multimodal data, and real-time difficulties covered in the research.
intraoperative feedback systems are necessary for future It also examines recent advancements in neural networks
developments. and the function of massive datasets in building reliable
models. Improving model interpretability, cutting energy
5. Deep Learning in Computer Vision: A Critical Review of use, and fusing AI with edge computing for real-time
Emerging Techniques and Application Scenarios applications are the main areas of future study. This review
offers insightful information on new developments in
The study examines the development of deep learning (DL) computer vision technology and trends.
for computer vision (CV), classifying its phases from 2012
8. Deep Learning for Consumer Devices and Services: REFERENCES
Pushing the Limits for Machine Learning, Artificial
[1] V. Mittal and B. Bhushan, "Accelerated Computer Vision Inference
Intelligence, and Computer Vision with AI on the Edge," 2020 IEEE 9th International Conference on
Communication Systems and Network Technologies (CSNT),
The quick developments in deep learning, especially in Gwalior, India, 2020, pp. 55-60, doi:
10.1109/CSNT48778.2020.9115770.
convolutional neural network (CNN) training, are covered [2] Gollapudi, S. (2019). Artificial Intelligence and Computer Vision. In:
in the study. The availability of bigger datasets and the Learn Computer Vision Using OpenCV. Apress, Berkeley, CA.
development of new graphics processing technology are https://doi.org/10.1007/978-1-4842-4261-2_1
credited with this acceleration of advancement. The study [3] M. Hussain, "When, Where, and Which?: Navigating the Intersection
of Computer Vision and Generative AI for Strategic Business
highlights deep learning's applications in computer vision, Integration," in IEEE Access, vol. 11, pp. 127202-127215, 2023, doi:
machine learning, and artificial intelligence, as well as its 10.1109/ACCESS.2023.3332468.
revolutionary effects on consumer goods and services. In [4] Quero, G.; Mascagni, P.; Kolbinger, F.R.; Fiorillo, C.; De Sio, D.;
order to push the limits of what these gadgets and services Longo, F.; Schena, C.A.; Laterza, V.; Rosa, F.; Menghi, R.; et al.
Artificial Intelligence in Colorectal Cancer Surgery: Present and
can accomplish, it also examines the opportunities and Future Perspectives. Cancers 2022, 14, 3803.
difficulties associated with incorporating deep learning into https://doi.org/10.3390/cancers14153803
consumer electronics. [5] Junyi Chai, Hao Zeng, Anming Li, Eric W.T. Ngai, Deep learning in
computer vision: A critical review of emerging techniques and
application scenarios, Machine Learning with Applications, Volume
9. Deep Learning-Enabled Medical Computer Vision 6, 2021, 100134, ISSN 2666-8270,
https://doi.org/10.1016/j.mlwa.2021.100134.
[6] Xin Feng, Youni Jiang, Xuejiao Yang, Ming Du, Xin Li, Computer
The paper offers a thorough overview of the ways in which vision algorithms and hardware implementations: A survey,
medical image processing is being revolutionized by deep Integration, Volume 69, 2019, Pages 309-320, ISSN 0167-9260,
learning, namely convolutional neural networks (CNNs). https://doi.org/10.1016/j.vlsi.2019.07.005.
The authors highlight CNNs' capacity to attain expert-level [7] Victor Wiley et.al,Computer Vision and Image Processing: A Paper
Review, International Journal of Artificial Intelegence Research, Vol.
performance in tasks including illness detection, 2, No. 1, June 2018, pp. 28-36,ISSN:2579-
segmentation, and classification as they examine their use in 7298,https://doi.org/10.29099/ijair.v2i1.42
a variety of medical imaging modalities, such as radiology, [8] J. Lemley, S. Bazrafkan and P. Corcoran, "Deep Learning for
pathology, dermatology, and ophthalmology. The necessity Consumer Devices and Services: Pushing the limits for machine
learning, artificial intelligence, and computer vision," in IEEE
for sizable annotated datasets, model interpretability, and Consumer Electronics Magazine, vol. 6, no. 2, pp. 48-56, April 2017,
workflow integration are some of the difficulties in doi: 10.1109/MCE.2016.2640698.
implementing these systems in clinical settings that are [9] Esteva, A., Chou, K., Yeung, S. et al. Deep learning-enabled medical
covered in the paper. Future directions stress how crucial it computer vision. npj Digit. Med. 4, 5 (2021).
https://doi.org/10.1038/s41746-020-00376-2
is to create reliable, broadly applicable models and make [10] Quant Imaging Med Surg. 2021 Aug;11(8):3830-3853. doi:
sure that ethical issues are taken into account when 10.21037/qims-20-1151
implementing AI in healthcare.

10. What is new in computer vision and artificial

intelligence in medical image analysis applications

With an emphasis on applications in cardiology, cancer,

dermatology, neurodegenerative diseases, and other fields,
the article explores the expanding roles of artificial
intelligence (AI) and computer vision (CV) in medical
picture analysis. It showcases developments in disease
categorization, segmentation, and image improvement. In
addition to providing special opportunities, several imaging
techniques including CT, MRI, ultrasound, and microscope
also have drawbacks like noise, resolution, and contrast
restrictions. In a variety of fields, including cardiology,
oncology, and dermatology, it discusses the application of
deep learning for tasks including image augmentation,
segmentation, and illness classification. Notwithstanding
significant advancements, problems still exist, including
handling many imaging modalities, poor data quality, and
the requirement for strong AI models. The future
possibilities and challenges of AI in medical diagnostics are
also discussed in the article.