The rapid surge in video content across domains such as surveillance, media, and online platforms has created an urgent need for intelligent systems that can automatically generate concise video summaries. Manual review of lengthy footage is often inefficient and impractical in time-sensitive scenarios. This paper proposes a real-time video summarization framework powered by deep learning. The system lever- ages ResNet-based Convolutional Neural Networks to extract key visual features, Bidirectional Long Short-Term Memory (BiLSTM) networks to model temporal dependencies, and an attention mechanism to identify the most informative frames. Evaluations were conducted on standard datasets such as SumMe and TVSum, both containing human-annotated summaries. Experimental results show that the proposed model delivers accurate and contextually coherent video summaries with minimal latency, making it suitable for real-time applications like surveillance monitoring, video search, and content indexing