Detection of False Synchronization of Stereo Image Transmission Using a Convolutional Neural Network

Abstract

The subject of the work described in this article is the detection of false synchronization in the transmission of digital stereo images. Until now, the synchronization problem was solved by using start triggers in the recording. Our proposal checks the discrepancy between the received pairs of images, which allows you to detect delays in transferring images between the left camera and the right camera. For this purpose, a deep network is used to classify the analyzed pairs of images into five classes: MuchFaster, Faster, Regular, Slower, and MuchSlower. As can be seen as a result of the conducted work, satisfactory research results were obtained as the correct classification. A high percentage of average probability in individual classes also indicates a high degree of certainty as to the correctness of the results. An author’s base of colorful stereo images in the number of 3070 pairs is used for the research.