Deep Learning-Driven Real-Time Video Summarization with Temporal Modeling and Attention Mechanism

Abstract

Developments in video content across various platforms require new advanced techniques to efficiently manage and streamline vast video data access. Using Recurrent Neural Networks (RNNs) and Bidirectional Long Short-Term Memory (BiLSTMs) together with attention mechanisms the paper addresses real-time video summarization. The developed system demonstrates functionality to detect essential keyframes while maintaining control over time-based sequence relations in video content. Video summaries created by humans serve together with raw video content to enable the model to discover vital visual data and contextual associations. Standards measuring effectiveness include BLEU and ROUGE which help assure both clear and coherent results from generated summaries. The proposed method demonstrates real-time summary generation accuracy for different video types based on user preferences which makes it a strong practical automatic video summarization technique.