Moran, Maira, Aura Conci, and Ángel Sánchez. "Automatic Detection of Knives in Complex Scenes." ICT Applications for Smart Cities. Springer, Cham, 2022. 57-77.
Smart Cities use a variety of Information and Communication Technologies (ICT) and databases to improve the efficiency and efficacy of city services. Security is one of the main topics of interest in this context. The increase in crime rates demands the development of new solutions for detecting possible violent situations. Video surveillance (CCTV) cameras can provide a large amount of valuable information contained in images which can be difficult to be analyzed by humans in an efficient form. Identifying and classifying weapons in such images is a challenging problem that can be driven by the application of Deep Learning techniques. Object detection algorithms, especially advanced Machine Learning ones, have demonstrated impressive results in a wide range of applications. However, they can fail in certain application scenarios. This work describes a novel proposal for knife detection in complex images. This is a challenging problem due to the multiple variabilities of these objects in scenes (i.e., changing shapes, sizes and illumination conditions, among others), which can negatively impact the performance of mentioned algorithms. Our approach analyzed the combination two super-resolution techniques (as a preprocessing stage) with one object detection network to effectively solve the considered problem. The results of our experiments show that the proposed methodology can produce better results when detecting small objects having reflecting surfaces (i.e., knives) in scenes. Moreover, the approach could be adapted for surveillance applications that need real-time detection of knives in places monitored by cameras.