The condition of road infrastructure is dynamic, constantly changing, and vital to ensure the safety and economic efficiency of an urban area, but current manual inspection techniques are subjective, labour intensive and time-consuming. This paper introduces an Automated Pothole Detection and Severity Mapping system that proposes using real time measurement to address early objective measurement of the pothole condition of the road surface. This monitoring approach takes a sensor-fusion point of view by incorporating visual monitoring via Computer Vision (CV) with physical monitoring via Inertial Measurement Unit (IMU). It applies a model that can detect fast speed of object moving in order to detect deformation in the surface and a model that applies a synchronized Accelerometer to detect vertical movement to determine the depth of the pothole. The hazards that are detected are classified into three different levels of severity (Minor, Moderate and Critical) and are automatically uploaded to a Geographic Information System (GIS) dashboard using GPS co-ordinates. Results from experimentation showed that false positives (like shadows or oil spills) can be greatly reduced by using visual and physical data simultaneously as against vision only. An automated, scalable solution to assist municipalities prioritize road repair, more effectively utilize resources, and enhance commuter safety.

Keywords : Automated Road inspection, Pothole detection, Deep learning, Severity mapping, IoT sensors, GIS.

Authors : M. Dharani , Ramesh Samudrala , Sk. Rukshana Begum

Title : Automated Pothole Detection And Severity Mapping Using Geographic Information System

Volume/Issue : 2026;3(1 ( January - March ))

Page No : 27 - 33

Even this type of long-haul driving occurs on highways, resulting in numerous crashes on its part. Nearly all of the latest tools monitor the face and/or man body signals individually, but that isn't always efficient when the setting modifications. However, what this study does instead is to create an integrated system using camera footage alongside real-time health sensors for the purpose of more readily identifying signs of tiredness. Cameras register eyes closed, frequency of blinks, mouth opening during a yawn, as well as head-directing position. Oh, and don't forget that heartbeat changes, blood flow adjustments and sweating responses are measured as well. Patterns extracted from images are fed into layers of neurons that can detect shapes, and another network of neurons learns how time unfolds over the course of seconds, to gauge drowsiness. Three phases emerge - wake up, declining attention, collapse; not assumptions, but multi-layered conclusions. Results are not caused by individual clues but filamentous. The timing shifts help maintain accuracy but when there is sudden change in light or rapid movement of lights or slight change in bodies, the setups simultaneously check eye signs as well as body stress signs. Computations are performed locally, reducing delays and warnings become available rapidly even in low bandwidth situations, right inside the device. According to tests, this comb of inputs is able to detect tiredness 95% of the time, outperforming the performance of previous methods that monitored one type of input signal. Rather than merely giving out static pop-up alarms, it creates a ‘real-time attention score' for drivers, which allows cars to determine what alerts to pay attention to at the moment. It is designed to be expanding and requires low electricity consumption and can propel future vehicles, smart transformers and more intelligent roads.

Keywords : Driver Fatigue Detection, Computer Vision, Biometric Monitoring, CNN, LSTM, Edge Computing.

Authors : Bharat Kumar Chigilipalli , VenkataRamana Guntreddi

Title : Multi - Modal Driver Fatigue Detection Using Computer Vision And Biometric Sensor Fusion

Volume/Issue : 2026;3(1 ( January - March ))

Page No : 19 - 26

Image captioning is one of the most important domains in human-computer interaction that generates a description of the visual information. In this study, the need for large-scale studies is motivated to investigate the benefits of deep learning techniques including the combination of CNN and LSTM models on image captioning. The proposed approach extracts visual features by CNN from the diverse images and generates coherent and contextually appropriate captions using the LSTM model from the Flickr8K database, which consists of various images with respective captions. The combination of CNN and LSTM layers is used to capture the spatial and temporal features, which improves the accuracy and relevance of the generated text descriptions. In the proposed method, the combination of CNN and LSTM layers allows for the efficient extraction of spatial and temporal information, significantly improving the accuracy and relevance of generated text descriptions. The purpose of this study is to measure the performance of CNN-LSTM model for image captioning, with the aim of enhancing visual understanding and boosting developments in natural language processing.

Keywords : Image Captioning, Deep Learning, CNNs, LSTM, Flickr8K Dataset.

Authors : Yegireddi Ramesh

Title : Image Captioning Using OpenCV, CNN and LSTM

Volume/Issue : 2026;3(1 ( January - March ))

Page No : 13 - 18

Epilepsy among children is one of the most common neurological conditions with a need for ongoing monitoring for prompt clinical intervention. Using conventional cloud based systems can lead to time delays and be problematic for privacy reasons and therefore are inappropriate for providing real-time pediatric care. A real time monitoring framework deployed in resource constrained edge devices is proposed in this paper. The architecture comprises of wearable EEG acquisition, bandpass, PCA-based feature extraction and then optimises the lightweight Temporal Convolutional Network (TCN) for deployment on TensorFlow Lite using model pruning and quantization. A classification accuracy of 97.6%, 70.8% lower than the cloud rely CNN approach, AUC of 0.987, and inference latency of 38ms are achieved with only 95mW being consumed during the experiment with the pediatric EEG benchmark. This finding shows that such on-device and privacy-preserving low latency seizure monitoring is possible for young patients.

Keywords : Pediatric Epilepsy, EEG, Edge Computing, TinyML, TCN, real-time seizure detection; wearable systems

Authors : M. Dharani , Nasreen Sultana Quadri , T. Venkata Krishnamoorthy

Title : Real-Time Paediatric Seizure Monitoring on Edge Devices Using Lightweight Temporal Convolutional Networks

Volume/Issue : 2026;3(1 ( January - March ))

Page No : 8 - 12

Machine learning systems in the brain signal processing field of electroencephalography (EEG) have established a good performance in neural signal analysis, emotion recognition systems, seizure detection and brain-computer interface (BCI). One major difficulty that needs to be addressed in real-world applications is that models derived from one EEG dataset will generally suffer a large drop in performance if tested on a different one, as a result of differences in equipment participating in the recordings, electrode placement, subject and demographic differences, environmental noise and individual subject variability. This paper introduces a new hybrid cross-dataset quantum domain adaptation (QDA) framework that combines covariance-based feature extraction, manifold projection, adversarial maximum mean discrepancy (MMD) domain alignment and parametric quantum kernel learning with parameterized quantum circuits. The experimental results validate that the proposed framework can attain 97.8% classification accuracy, which exceeds the accuracy of conventional CNN (89.2%), CNN-QSVM (92.4%), and Transformer (94.6%) models. It also enhances accuracy across all the metrics like precision, recall, and F1-score.

Keywords : EEG, Quantum Machine Learning, Transfer Learning, QSVM, Brain-Computer Interface.

Authors : Sateesh Gudla

Title : Cross-Dataset Domain Adaptation for Quantum EEG Classification Models

Volume/Issue : 2026;3(1 ( January - March ))

Page No : 1 - 7