Neural Network Video Streaming

In the era of big data, the field of computer vision is experiencing rapid growth. Computer vision allows solving various tasks, one of which is tracking moving objects in the video stream.The main requirements for object tracking systems in video stream are: high accuracy of objectmovement detection and the ability to process sequence of frames in real time (more than 120 frames per second).This paper presents an intelligent system for tracking the identification numbers of railway numbers in a video stream. The developed system is not pre-trained for tracking any specific objects, but allows to determine the target object at start-up. The purpose of the study was to develop an algorithm for tracking objects in the video stream, invariant with respect to the properties of the object (color, shape, etc.) and the environment (eg, lighting). Therefore, recognition is performed on the basis of video surveillance of four cameras, to obtain a single image. Also the algorithm takes into account the possible recovery of the tracked number of poor quality and additional information in the image.