A Novel Steady and Redundant Frames Removal Performance Analysis of Video Background Subtraction Algorithms with YOLOx Models

One of the prime research areas today is video summarisation, where the short summary of the input video is generated. The output video content is based on the object of interest. It also saves the time of the viewer; they won't need to view the entire long duration video. To do this process faster, the redundant or steady frames need to be removed first. Background subtraction is one of the techniques used to identify moving objects by calculating the difference between previous and current frame. In our research paper, we have used KNN, MOG2 and CNT for the BGS methods. We have also used different YOLO model like YOLOv3, YOLOv5, YOLOv7 and YOLOv8 to generate annotations file and use them for measuring the performance of the BS algorithms. These methods are implemented using parallel processing approach to process more than one video at a time. The performance analysis of these methods is done based on the time of execution and precision. The experiment result shows that the KNN with YOLOv7 perform better and parallel processing approach speedup the execution.