Stopping Method for Iterative Scene Identification in an Image Stream

As part of solving a set of computer vision problems when analyzing video streams, a fairly popular task is to effectively stop the process of identifying scenes or objects. Arresters are one of the main problems in systems that implement online computer vision. This is explained by the fact that period of time spent obtaining results in video analytics systems is no less important than the reliability of the analysis results. In this regard, the conducted research focused on the problem of stopping when identifying scenes in a video stream. This requires detailed stopping algorithms for identifying scenes in video streams with a decision to arrest iterations, taking into account that it is not possible to generally assess the accuracy of the identification result. In this work, a list of properties of monotonic stopping functions was compiled, and an original algorithmic technique for solving stopping problems was proposed. The testing results demonstrated the high efficiency of the proposed method for identifying the text part of documents.