INTELLIGENT PROCESSING OF IMAGES FROM VISION SYSTEMS AT STAGES OF AIRCRAFT LANDING

Since 2016, the appearance of the 4th generation intelligent technical vision system has begun to take shape based on the use of modern highly sensitive sensors of different spectral ranges and physical nature. This research solves the problem of determining the location of the runway and its elements on images of an underlying surface from the technical vision system at the stages of approach and landing. The proposed method includes separate neural network architectures that solve the following subtasks: the runway detection, the runway’s threshold detection, the symbol detection (the runway number) and clarification of additional the runway parameters such as: a position of its corner points, a position of the runway’s threshold, a determination of the degree of reliability and a determination of an inclination angles of the runway sidelines relative to the horizon. The developed method is based on such neural network architectures as YOLO8n-p6 for the task of object detection, MnasNet and RegNet for the task of regression of 4 corner points of an object, as well as on the architecture of multivalued classification for runway number recognition. To neutralize effect of possible difficult weather conditions during the landing process it’s proposed to track the runway corner points using Kalman filter. The results of a simulation modelling of a determination of the runway location and its elements based on processing public videos of aircraft’s landing stage are presented.