Approach to Building an Intelligent System for Detecting Anomalous Objects in the Railway Zone Using Real-Time Video Stream

the article discusses an approach to building an intelligent system for detecting anomalous objects in the railway zone using real-time video stream. The relevance of the research is driven by the need to improve the efficiency and objectivity of monitoring the state of railway infrastructure and rolling stock, as well as to reduce the influence of the human factor in the analysis of visual information. Unlike classical detection tasks for predefined objects, this work focuses on identifying deviations from the normal state of the observed scene, which allows covering a wide class of potentially dangerous and technologically abnormal situations. A formalised problem statement for anomalous object detection is proposed, the main elements of the structural and functional diagram of the system are defined, and an algorithm for video stream processing is described, including the selection of the controlled zone, feature extraction, computation of an anomaly score, classification, and temporal verification of detected events. It is shown that the proposed approach can be applied both when using unmanned aerial systems and with stationary camera placement. Special attention is paid to the requirements for real-time system operation and the formation of an integral decision-making criterion. The conclusion is drawn about the promise of the proposed approach for subsequent experimental verification and integration into intelligent monitoring systems for railway transport facilities.