Research on Video Keyframe Extraction Method Based on Action Analysis

Abstract

In the field of behavioral analysis, dividing keyframes based on video images to extract the motion information is one of the more common preprocessing methods, which can preserve as much semantic information as possible while compressing the data. In this paper, we propose a method for dividing video keyframes based on the analysis of human motion in video, which designs feature vectors based on human motion characteristics and gives suggestions for keyframe division by analyzing the change of quantitative values. It is experimentally confirmed that the method is less data-dependent, less time-consuming to engineer, more lightweight in the design of feature vectors and closer to the expected division results while ensuring relatively stable and reliable recognition results for downstream tasks, which provides more reliable antecedent support for the content analysis work of human motion videos.