Unauthorized runway entry warning system at airports using deep learning method

A large number of people make extensive use of airports for transportation and travel purposes. Consequently, both the high volume of individuals and the intensity of traffic can lead to the emergence of potential criminal activities within airport premises. In response, advancements in technology and the implementation of modern security management strategies have aimed to enhance safety measures and reduce the likelihood of accidents and attacks. Security systems are employed to monitor terminals and passengers, enabling the identification of possible vulnerabilities. Through such systems, incidents occurring or likely to occur at airports can be detected in advance. Around airport perimeters, various applications are implemented, such as monitoring and preventing border breaches or the intrusion of living beings onto the runway during aircraft takeoff or landing. This study focuses on early warning systems, which are an important issue for airport security. Image processing techniques have been utilized with a particular focus on the detection of humans, objects, and animals, in order to prevent potential threats around airport environments. The study consists of two applications. In this way, the results of the study will be observed according to class differences. In the first application, the number of classes was kept little and the success level of the study was observed. In this application, violation scenarios were tested with imaginary boundaries determined by object detection and classification with human, object (tree) and animal (cat) images. Based on the test dataset generated through the training of these classes, the success rates were obtained as 99% for trees, 94% for humans, and 100% for cats. In the second implementation, the number of classes was increased. In addition to humans, objects (trees), and animals (cats), images of cows and goats were also included. Object detection and classification were again performed, and virtual boundary intrusion scenarios were tested. According to the results obtained from the test dataset, which was formed using images of all these categories, the success rates were determined as 91% for trees, 86% for humans, 84% for cats, 89% for cows, and 75% for goats.