Atul Kumar Mishra
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Atul Kumar Mishra has a total of 6 co-authors. Here are the top co-authors by number of shared publications:
- Koratagere Sreelakshmi
Department of Electronics and Telecommunication Engineering, RV College of Engineering, Bengaluru, India
1 shared publication - Nishant Vasantkumar Hegde
Department of Computer Science and Engineering, RV College of Engineering, Bengaluru, India
1 shared publication - Pratiba Deenadhayalan
Department of Computer Science and Engineering, RV College of Engineering, Bengaluru, India
1 shared publication - Ramakanth Kumar Pattar
Department of Computer Science and Engineering, R. V. College of Engineering, Bengaluru, India
1 shared publication - Saksham Gupta
Department of Computer Science and Engineering, RV College of Engineering, Bengaluru, India
1 shared publication T. Shankar
Center for Computer Vision Research, RV College of Engineering, Bengaluru, India
1 shared publication
Latest Publications
- Open Access
- 27 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 64277-64301
An Integrated Dual-Module Computer Vision System for IV Drip Rate and Fluid Level Monitoring
Manual monitoring of intravenous (IV) therapy is labor-intensive and susceptible to human error, posing risks to patient safety. Existing computer vision-based approaches have addressed either drip rate estimation or fluid level monitoring in isolation, limiting their clinical applicability. This work presents and validates a dual-module computer vision framework for a practical, automated IV monitoring system that simultaneously addresses both tasks. For the critical task of drip rate estimation, we conduct a comparative study of two distinct deep learning paradigms on a diverse custom dataset of 3,458 images, evaluated on a strictly held-out test set: a state-of-the-art YOLOv8 object detector, achieving mAP0.50 of 95.34%, and a novel heatmap-based point-event regression network (ResNet-18 encoder with convolutional decoder), achieving a frame-wise accuracy of 88.44% and recall of 95.16%. The second module leverages a bespoke convolutional neural network (CNN) for fluid level classification, reaching a test accuracy of 95.26% on a public benchmark dataset. Finally, we demonstrate a proof-of-concept prototype that successfully integrates both modules into a unified, automated video-based application, validated on pre-recorded clinical video footage. End-to-end counting evaluation on a 7-video set spanning drip rates from 17 to 94 dpm shows the heatmap-based counter achieves 91.0% mean counting accuracy versus 77.5% for the YOLOv8-based counter, with the heatmap demonstrating markedly greater robustness at elevated infusion rates. By validating high-performance solutions for both monitoring tasks and demonstrating a clear path to integration, this study establishes a comprehensive framework for a holistic IV infusion monitoring system with the potential to enhance patient safety and reduce the burden on clinical staff.
- 27 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 64277-64301
Open AccessAn Integrated Dual-Module Computer Vision System for IV Drip Rate and Fluid Level Monitoring
Manual monitoring of intravenous (IV) therapy is labor-intensive and susceptible to human error, posing risks to patient safety. Existing computer vision-based approaches have addressed either drip rate estimation or fluid level monitoring in isolation, limiting their clinical applicability. This work presents and validates a dual-module computer vision framework for a practical, automated IV monitoring system that simultaneously addresses both tasks. For the critical task of drip rate estimation, we conduct a comparative study of two distinct deep learning paradigms on a diverse custom dataset of 3,458 images, evaluated on a strictly held-out test set: a state-of-the-art YOLOv8 object detector, achieving mAP0.50 of 95.34%, and a novel heatmap-based point-event regression network (ResNet-18 encoder with convolutional decoder), achieving a frame-wise accuracy of 88.44% and recall of 95.16%. The second module leverages a bespoke convolutional neural network (CNN) for fluid level classification, reaching a test accuracy of 95.26% on a public benchmark dataset. Finally, we demonstrate a proof-of-concept prototype that successfully integrates both modules into a unified, automated video-based application, validated on pre-recorded clinical video footage. End-to-end counting evaluation on a 7-video set spanning drip rates from 17 to 94 dpm shows the heatmap-based counter achieves 91.0% mean counting accuracy versus 77.5% for the YOLOv8-based counter, with the heatmap demonstrating markedly greater robustness at elevated infusion rates. By validating high-performance solutions for both monitoring tasks and demonstrating a clear path to integration, this study establishes a comprehensive framework for a holistic IV infusion monitoring system with the potential to enhance patient safety and reduce the burden on clinical staff.