VOLUME 14, ISSUE 8, AUGUST 2026

Abstract: The automation of vehicle damage assessment has emerged as a cornerstone requirement for modernizing the automotive insurance, rental, and fleet management industries. Traditional inspection paradigms rely heavily on manual labor, leading to subjectivity, slow processing times, and high administrative overhead. This research presents a novel, comprehensive AI-powered vehicle damage detection system that mitigates these challenges through advanced deep learning techniques. By leveraging the state-of-the-art YOLOv8 architecture, our system enables real-time capability in identifying, localizing, and classifying multiple classes of superficial and structural vehicular damages. This paper delineates the end-to-end framework: starting from robust dataset preprocessing pipelines and moving toward an intricate architectural design composed of a FastAPI backend proxy and a React.js client interface. Furthermore, we provide an extensive mathematical formulation of the underlying neural network operations, detailing the equations driving spatial convolutions, Intersection over Union (IoU) metrics, and custom Focal Loss computations. Empirical evaluations demonstrate an exceptional Mean Average Precision (mAP@0.5) of 0.942, establishing the viability of this solution for scalable, production-grade deployment environments. This work extensively validates the integration of micro-services with edge-compatible AI vision models, paving the way for fully autonomous vehicle inspection checkpoints.

Abstract: Rising incidents of weapon-related violence in public spaces have intensified demand for automated surveillance systems capable of real-time weapon identification. This paper presents a complete end-to-end weapon detection and alerting pipeline built on the YOLOv8 nano architecture. The system identifies three weapon categories — knives, pistols, and rifles — from live camera streams, webcam feeds, and static images. To address the well-known false-positive problem in weapon detection, we introduce a structured hard-negative mining strategy: background images sourced from diverse indoor environments (offices, homes, classrooms, public spaces) are automatically screened by the model; frames that generate spurious detections are harvested and re-introduced into the training corpus with empty labels. The dataset, comprising approximately 11,500 annotated images across three weapon classes, is further augmented with 1,500+ negative samples. After five iterative fine-tuning rounds with mosaic, mixup, and copy- paste augmentation, the final model achieves an overall mAP@50 of 0.727 (knife 0.75, pistol 0.55, rifle 0.88). An accompanying alert subsystem enforces temporal debouncing — requiring N consecutive high-confidence frames before triggering — and dispatches annotated evidence images to operators via Telegram or configurable HTTPS webhooks. A Flask-based local dashboard provides a browser interface for image analysis, video processing, and live webcam monitoring. Evaluation results, qualitative analyses, known limitations (notably a pistol precision of approximately 0.41), and recommendations for production deployment are discussed in detail.

Abstract: Traffic monitoring plays a vital role in improving road safety and transportation management. This paper presents a Vehicle Tracking and Traffic Analytics System that utilizes YOLOv11 for real-time vehicle detection and ByteTrack for multi-object tracking. The system estimates vehicle speed, detects overspeeding vehicles, analyzes traffic movement through predefined regions of interest, and generates traffic density heatmaps. It processes highway surveillance videos to provide accurate vehicle tracking and real-time traffic insights. The proposed system offers an efficient and scalable solution for intelligent transportation systems, traffic surveillance, and highway monitoring.

Keywords: Vehicle Tracking, Traffic Analytics, YOLOv11, ByteTrack, Speed Estimation, Computer Vision

Abstract: Efficient image compression is essential for minimizing storage requirements and reducing transmission bandwidth while preserving the visual quality of reconstructed images. Among the various compression techniques there is a Differential Pulse Code Modulation (DPCM) represents an effective predictive compression approach has gained considerable attention so because of its predictive coding mechanism, which utilizes the spatial dependency among the adjacent pixels by encoding prediction errors rather than absolute pixel values. This approach effectively reduces data redundancy by maintaining low computational complexity, making it well suited for real-time image computing applications.
This paper presents a detailed review of DPCM-based image compression methods, including their operating principles, prediction techniques, hardware architectures, and performance evaluation. It also discusses recent improvements in adaptive and hybrid DPCM approaches that enhance compression efficiency and reconstructed image quality. Furthermore, the study compares existing research based on compression performance, hardware resource utilization, implementation complexity, and processing speed. Finally, current research challenges, emerging trends, and future opportunities involving intelligent prediction methods, machine learning, and next-generation image coding techniques are highlighted. The review concludes that DPCM continues to be a practical, efficient, and hardware-friendly approach for real-time image compression applications.

Keywords: Differential Pulse Code Modulation (DPCM), comparative analysis, Image compression, Wavelet Transform, Biomedical Signal Processing, Real-Time Signal Processing, Sustainable Healthcare, Wearable Healthcare.

Abstract: Cardiovascular diseases are a major cause of death and illness worldwide. This is why accurate Electrocardiogram (ECG) monitoring is so important for finding and treating these conditions early. Unfortunately, ECG signals can be unclear because of things like baseline drift, power line noise, movement, electrode issues, and electrical signals from muscles. This makes it harder to get a reliable diagnosis. To fix this, many methods have been developed to clean up ECG signals. These include digital filters, wavelet transforms, adaptive filters, deep learning, and empirical mode decomposition. When it comes to hardware, Field Programmable Gate Arrays (FPGAs) are a good choice for cleaning ECG signals in real time. They can process data in parallel, have low delay, can be reprogrammed, and use energy efficiently. This paper looks at recent methods for cleaning ECG signals using FPGAs. We compare their algorithms, how they are optimized for hardware, how complex they are to implement, what resources they use, and how well they remove noise. We also discuss how FPGAs are being used with Artificial Intelligence (AI), Edge Computing, the Internet of Things (IoT), and wearable health devices. Finally, we talk about the current research challenges and what needs to be done in the future to create ECG denoising systems that are scalable, use little power, and perform well.

Keywords: Electrocardiogram (ECG), FPGA, ECG Denoising, Wavelet Transform, Biomedical Signal Processing, Real- Time Signal Processing, Sustainable Healthcare, Wearable Healthcare.

Abstract: Predictive maintenance (PdM) has emerged as a key application of artificial intelligence, machine learning, and data analytics in modern industry. Unlike preventive maintenance, predictive maintenance utilizes historical and real-time sensor data to identify equipment degradation and estimate potential failures before they occur. Python has become one of the most widely adopted programming languages for predictive maintenance due to its extensive ecosystem of scientific and machine learning libraries such as NumPy, Pandas, Scikit-learn, Matplotlib, and Seaborn. Among clustering validation techniques, silhouette analysis plays a significant role in determining the optimal number of clusters and evaluating the quality of clustering results. This review paper presents a comprehensive overview of predictive maintenance using Python with silhouette analysis. It discusses maintenance strategies, machine learning approaches, clustering techniques, silhouette coefficient interpretation, Python-based implementation frameworks, publicly available datasets, and recent research contributions. The paper also highlights advantages, limitations, and future research directions in predictive maintenance systems employing unsupervised learning and silhouette-based cluster validation.

Keywords: Predictive Maintenance, Python, Silhouette Analysis, Machine Learning, K-Means Clustering, Unsupervised Learning, Industrial Analytics, Condition Monitoring.

Abstract: Quick-commerce fulfillment centers (dark stores) operate under strict order-to-delivery timelines, making manual quality assessment of perishable fruits and vegetables highly inefficient and error-prone. This research paper presents a robust, universal Computer Vision (CV) system integrated with a non-linear regression model for automated freshness classification, rot segmentation, and quantitative stock loss estimation. The system utilizes HSV color space analysis and edge gradient filters to segment the produce body, eliminate complex backdrops (including wooden tables and studio backgrounds), and isolate localized rot spots. Furthermore, we introduce a quantitative stock estimation model that maps pixel density ratios to physical weight, allowing operators to automatically determine the exact quantity (in kilograms) of rotten produce within a bulk batch. Tested across tomato, banana, capsicum, and spinach images, the system successfully eliminates false positives from healthy features like green stems and natural yellow/red skin variations, achieving a classification accuracy of 96.8%.

Keywords: Computer Vision, Freshness Detection, Stock Quality Analysis, Rot Spot Masking, Image Segmentation, Quantity Estimation, Quick Commerce.

Abstract: Community pharmacies and small dispensaries in India largely continue to track stock, batch numbers, and expiry dates on paper registers or in disconnected spreadsheets, a practice that leaves them exposed to stock-outs, undetected expired inventory, and slow reconciliation. This paper presents Smart Pharmacy AI, a full-stack web application that digitizes medicine inventory management through a Flask and SQLAlchemy REST API backed by a five-table relational data model, and a React (Vite, Tailwind CSS) single-page front end. The system exposes CRUD endpoints for medicine records and a dashboard endpoint that aggregates total stock, low-stock counts, and expired-item counts directly from the database using SQL aggregate functions. Beyond the operational CRUD and analytics layer, the system is architected with dedicated data models and route modules (Scan History, OCR Record, Inventory History, Chat History) that are designed to host a computer-vision medicine-package detector, an OCR-based batch/expiry-date extractor, and a conversational assistant as they are progressively integrated. We describe the system architecture, the relational schema, the implemented REST endpoints, the front-end design, the testing procedure followed, security and data-integrity considerations, a comparison against manual and generic point-of-sale alternatives, and the current implementation status of every module, before outlining the machine-learning components planned for the computer- vision and OCR subsystems and the corresponding development roadmap.

Keywords: Pharmacy Inventory Management, Flask REST API, SQLAlchemy ORM, React.js, Single-Page Application, Computer Vision, Optical Character Recognition, Expiry Tracking, AI Chatbot, Full-Stack Web Development.

Abstract: Person re-identification (Re-ID) across camera networks remains sensitive to illumination variation, since a single global appearance encoding is rarely robust to both bright and poorly lit capture conditions. This paper presents an adaptive illumination-robust Re-ID pipeline that estimates a continuous per-crop illumination score, generates four parallel enhanced views of each detected person (raw, CLAHE, gamma-corrected, and multi-scale Retinex), extracts an OSNet embedding for each view, and fuses the four embeddings using illumination-conditioned weights before performing cosine-similarity gallery retrieval via FAISS. The system is evaluated on Market-1501 under a stratified protocol comparing a Normal split against a synthetically generated Lowlight split, and comparing the proposed adaptive fusion against a static/average-fusion ablation baseline. Results show Rank-1 accuracy saturated at 98-99.6% across all four conditions, while mean Average Precision (mAP) is more discriminative, falling from approximately 71% on the Normal split to 44-46% on the Lowlight split. Contrary to the design hypothesis, the heuristic adaptive fusion does not outperform static average fusion in this evaluation, and static fusion is marginally better under Lowlight conditions. These findings are presented transparently as evidence that the current rule-based fusion weighting requires learning-based replacement, motivating a concrete direction for future work rather than an unqualified robustness claim.

Keywords: person re-identification, illumination robustness, adaptive feature fusion, OSNet, YOLOv8, Retinex, CLAHE, FAISS, Market-1501