Stabilization of a Dynamic Video Stream Taking Into Account Vehicle Movement Characteristics

This paper proposes a method for stabilization of a video stream received by a camera that moves with the vehicle. The proposed image stabilization is carried out taking into account the determined oscillation parameters of the vehicle itself. The main types of stabilization and the main measures of image quality assessment are considered. A model of image blurring based on factors that affect the final value of the blur is presented. The obtained images were tested before and after stabilization on the example of an object recognition task with an algorithm using the YOLO neural network. To assess image quality, it is proposed to use a measure of object image blur, which contains useful information about the state and behavior of the object. It has been determined experimentally that the method proposed in this article for stabilizing images of a moving object based on the use of information about its oscillations increases the accuracy of its recognition.