An Effective Object Detection Algorithm for High Resolution Video by Using Convolutional Neural Network

Abstract

In this paper, an algorithm to detect small objects more accurately in high resolution video is proposed. For this task, an analysis of state-of-the-art algorithms in application to high resolution video processing, which can be implemented into modern surveillance systems is performed. The algorithm is based on CNN in application to high resolution video processing and it consists of the following steps: each video frame is divided into overlapping blocks; object detection in each block with CNN YOLO is performed; post processing for extracted objects in each block is done and merging neighbor regions with the same class probabilities is performed. The proposed algorithm shows better results in application to small objects detection on high resolution video than famous YOLO algorithm.