Aerospace Instrument-Making Annotation << Back Object Detection and Tracking System With Optical Flow Correction Rastrepin D.E., Neusypin K.A., Khomutsky I.D. An algorithm for neural network detection and tracking of objects in real time with correction by optical fl ow is presented. The tracking task uses an extended Kalman fi lter with refi nement of the state vector by optical fl ow calculated by the Gunnar Farnebak method. The algorithm is resistant to changes in lighting, time of day and type of the sensor matrix. The results are confirmed by experimental data obtained when detecting objects in the Black Sea using a mobile optical-electronic system developed by JSC Central Research Institute CYCLONE. Keywords: Deep learning, convolutional neural networks, optical flow, weighted average method, extended Kalman filter, object detection and tracking. DOI: 10.25791/aviakosmos.2.2025.1464 Pp. 56-64.