10.64886/oajea.0102.004

Full Text Download

Abstract: Pattern recognition, computer vision, and image processing all are benefitted from hand-written alphabet recognition and categorization. A profusion of applications based on this domain have been created in the last few decades, such as sign identification, multilingual learning systems, and so on. This research shows how neural networks may be used to create a system that recognizes hand-written English alphabets in the air using hand gestures. Because of the acoustic similarities between the letters of the alphabet, this is a challenging undertaking to complete. The main problem is dealing with enormous different ways to write used by multiple peoples. There are a variety of alphabet-writing approaches in these complicated handwritten styles. The recognition of handwritten English alphabets has been the subject of several research studies. Several studies have been conducted on this subject, but none have proven effective in detecting English alphabets instantly moving your fingers in the breeze. Therefore, this article explains how to create an English Alphabet model of awareness that uses a Convolution Neural Network (CNN) to identify English alphabets based on hand motions the gap in the air. After a full analysis, this recommended approach achieved 93.08\% accurate responses over the EMNIST dataset.

10.64886/oajea.0102.006

Full Text Download

Abstract: Road traffic accidents and overspeeding remain critical public safety challenges worldwide, disproportionately affecting rapidly urbanizing regions with limited automated enforcement capacity. This paper presents an IoT-Enabled Speed and Accident Detection Platform that incorporates multi-object tracking and deep-learning based object recognition, and calibration-based speed estimation into a unified, real-time intelligent traffic monitoring framework. The proposed system employs YOLOv8 for accurate, high-speed vehicle detection; ByteTrack for consistent persistent identity assignment across consecutive video frames; and a pixel-to-real-world-distance calibration method for precise vehicle speed computation. The platform automatically classifies vehicle types, identifies overspeeding behavior against configurable speed thresholds, generates real-time visual alerts, counts traffic volume through a configurable virtual line-crossing mechanism, and supports future IoT and cloud-platform integration for accident detection, remote monitoring, and emergency response automation. Experimental evaluation on traffic video sequences recorded at an urban intersection in Dhaka, Bangladesh, shows a mean Average Precision (mAP@0.5) as 0.82, a Multi-Object Tracking Accuracy or MOTA of 0.75, an IDF1 of 0.79, and a speed estimation Mean Absolute Error or MAE as 3.2 km/h. The reconfigurable, scalable architecture positions this platform as a practical and cost-effective foundation for intelligent transportation systems (ITS) and smart city infrastructure.