FUNDAMENTALS AND METHODS OF OBJECT DETECTION IN VIDEO IMAGES USING NEURAL NETWORKS

Abstract

In this paper, small object detection or neural networks using multiple consecutive frames and speed detection consistency in video sequences can be seen. For example, by interpolating features between frames. This work introduces a new approach to perform online video object detection using two consecutive frames of video.

Methodological aspects of training neural networks to understand objects. In some cases, there are situations where the network cannot learn to understand the object. This happens when, at some stage of training, it becomes impossible to search for the legality between the training parameters and the results.

Experiments show that using a pre-frame improves performance over single-frame detectors, but using a distinct optical flow generally does not.