High-resolution aerial monitoring using DL for identifying abnormal activity based on visual patterns in drone videos

Abstract

Unmanned aerial vehicles (UAVs) and sophisticated deep learning (DL) models have made the application of artificial intelligence (AI) more popular. This has resulted in an increase in the number of attempts to improve high-resolution aerial monitoring using DL for identifying abnormal activity based on visual patterns in drone videos. The study introduces a one-class support vector machine (OC-SVM) oddity locator for low-altitude, limited-scope UAVs used for ethereal video surveillance. The primary goal is to improve UAV-based observation capabilities by identifying areas or things of interest without prior knowledge, hence improving tasks like queue control, vehicle following, and hazardous product identification. The framework makes use of OC-SVM because of its quick and lightweight setup, making it suitable for continuous operation on low-computational UAVs. It empowers the identification of several peculiarities necessary for low-elevation reconnaissance by using textural characteristics to recognise both large-scale and tiny structures. Examine the UAV mosaicking and change location (UMCD) dataset to demonstrate the effectiveness of the framework, which achieves excellent accuracy and outperforms traditional methods by about one fifth in a variety of metrics. The suggested model compares with current methods, demonstrating superior accuracy and performance in recognition of peculiarities. Evaluation metrics include F1-score, review, exactness, and accuracy. The model demonstrates that it always encounters an oddity with a review compromise of up to seven on ten, achieving complete accuracy.