School of Computer Science and Engineering (Jan – Mar 2023)
1. Goyal, A., Mandal, M., Hassija, V., Aloqaily, M., & Chamola, V. (2023b). Captionomaly: A Deep Learning Toolbox for Anomaly Captioning in Social Surveillance Systems. IEEE Transactions on Computational Social Systems, 1–9. https://doi.org/10.1109/tcss.2022.3230262
Abstract: Real-time video stream monitoring is gaining huge attention lately with an effort to fully automate this process. On the other hand, reporting can be a tedious task, requiring manual inspection of several hours of daily clippings. Errors are likely to occur because of the repetitive nature of the task causing mental strain on operators. There is a need for an automated system that is capable of real-time video stream monitoring in social systems and reporting them. In this article, we provide a tool aiming to automate the process of anomaly detection and reporting. We combine anomaly detection and video captioning models to create a pipeline for anomaly reporting in descriptive form. A new set of labels by creating descriptive captions for the videos collected from the UCF-Crime (University of Central Florida-Crime) dataset has been formulated. The anomaly detection model is trained on the UCF-Crime, and the captioning model is trained with the newly created labeled set UCF-Crime video description (UCFC-VD). The tool will be used for performing the combined task of anomaly detection and captioning. Automated anomaly captioning would be useful in the efficient reporting of video surveillance data in different social scenarios. Several testing and evaluation techniques were performed. Source code and dataset: https://github.com/Adit31/Captionomaly-Deep-Learning-Toolbox-for-Anomaly-Captioning
2. Mahapatra, A. K., Panda, N., & Pattanayak, B. K. (2023b). Quantized Salp Swarm Algorithm (QSSA) for optimal feature selection. International Journal of Information Technology, 15(2), 725–734. https://doi.org/10.1007/s41870-023-01161-6
Abstract: Metaheuristic algorithms are well-known and widely used strategies for tackling optimization issues. Each has advantages and limitations and is frequently combined with other algorithms to compensate for flaws. The basic Salp Swarm Algorithm (SSA) is simple to use and often effective when solving real-world optimization problems, but it can sometimes get stuck at local optima, leading to premature convergence. The main reasons for this are the poor population diversity, the lack of exploitable resources, and the exploration capabilities being insufficient. A modified SSA algorithm called quantized SSA (QSSA) is suggested to improve performance. The proposed method has incorporated a mathematical operator called the quantization operator into the basic SSA. The main goal of incorporating quantization operator is to improve population diversity and local usage, which can help in finding the solution space more effectively, thereby enabling faster convergence. The suggested QSSA approach is validated through IEEE-CEC-2014 Basic functions. Further, as an application, the same methodology is used to select the finest features from benchmark datasets while retaining accuracy and reducing neural network complexity.
3. Bandyopadhyay, A., Sarkar, A., Swain, S., Banik, D., Hassanien, A. E., Mallik, S., . . . Qin, H. (2023b). A Game-Theoretic Approach for Rendering Immersive Experiences in the Metaverse. Mathematics, 11(6), 1286. https://doi.org/10.3390/math11061286
Abstract: The metaverse is an upcoming computing paradigm aiming towards blending reality seamlessly with the artificially generated 3D worlds of deep cyberspace. This giant interactive mesh of three-dimensional reconstructed realms has recently received tremendous attention from both an academic and commercial point of view owing to the curiosity instilled by its vast possible use cases. Every virtual world in the metaverse is controlled and maintained by a virtual service provider (VSP). Interconnected clusters of LiDAR sensors act as a feeder network to these VSPs which then process the data and reconstruct the best quality immersive environment possible. These data can then beleveraged to provide users with highly targeted virtual services by building upon the concept of digital twins (DTs) representing digital analogs of real-world items owned by parties that create and establish the communication channels connecting the DTs to their real-world counterparts. Logically, DTs represent data on servers where postprocessing can be shared easily across VSPs, giving rise to new marketplaces and economic frontiers. This paper presents a dynamic and distributed framework to enable high-quality reconstructions based on incoming data streams from sensors as well as to allow for the optimal allocation of VSPs to users. The optimal synchronization intensity control problem between the available VSPs and the feeder network is modeled using a simultaneous differential game, while the allocation of VSPs to users is modeled using a preference-based game-theoretic approach, where the users give strict preferences over the available VSPs.
- Banik, D., Pal, S., Naskar, M. N. B., & Bandyopadhyay, A. (2022). Transformer Based Technique for High Resolution Image Restoration. https://doi.org/10.1109/ocit56763. 2022.00109
Abstract: Due to the difficulty of the problem, there have been many different approaches and algorithms developed to date. Defocus deblurring, single image motion deblurring, image deraining, and image denoising are some of the further branches of picture restoration. There are numerous algorithms that are appropriate in a variety of circumstances. For instance, gaussian noise like pepper and salt noise responds well to the median filter. The adaptive filter, on the other hand, performs particularly well in DSP and ANC applications. It has been discovered that methods based on deep neural networks, such as the restormer, can occasionally outperform earlier algorithms. We look for the best algorithm for image denoising, defocus deblurring, single image motion deblurring, and picture deraining in this work. Although restormer is shown to perform better than the others in the majority of circumstances, hybrid median filter is sometimes found to perform better than all of them combined.
- Chakraborty, S., Sambhavi, S., Panda, P., & Nandy, A. (2023b). An Ensemble Model for Gait Classification in Children and Adolescent with Cerebral Palsy: A Low-Cost Approach. In Lecture notes in networks and systems (pp. 73–83). Springer International Publishing. https://doi.org/10.1007/978-981-19-6525-8_7
Abstract: A fast and precise automatic gait diagnostic system is an urgent need for real-time clinical gait assessment. Existing machine intelligence-based systems to detect cerebral palsy gait have often ignored the crucial issue of performance and computation speed trade-off. This study, in a low-cost experimental setup, proposes an ensemble model by combining fast and deep neural networks. The proposed system demonstrates a competing result with an overall ≈82% of detection accuracy (sensitivity: ≈78%, specificity: ≈84%, and F1-score: ≈83%). Although the improvement in detection performance is marginal, the computation speed increased remarkably from state of the art. From the perspective of computation time and performance trade-off, the proposed model demonstrated to be competing.
1. Dr. Anjan Bandyopadhyay, Tunir Bhattacharya, Soma Debnath, Arup Roy, Debojoyti Banik, Prasant Kumar Pattanaik, “ Home Automation – Mechanism Design of Smart Appliances Application Number: 202331001257, Date of Filing: 4/01/202, Publish: 10/02/2023.
A study of different dialects implementation of Odia language for NEP compilation using Artificial Intelligence: Use Case for Kalinga Institute of Social Sciences
Dr. Manjusha Pandey
Project Director & Associate Professor at School of Computer Engineering (SCE), KIIT DU
Dr. Siddharth Swarup Rautaray
Co-Project Director & Associate Professor, School of Computer Engineering (SCE), KIIT DU
Dr. S.N. Mishra
Co-Project Director and Professor at KIIT School of Management.