Page 16 - TECH MAGAZINE CSE

AN Technical Magazine
VESHAN Maharaja Agrasen Institute of Technology

FACULTY
CORNER

Dr Ruchi Goel

Department of Computer Science &
Engineering

Explainable AI (XAI) in Video Summarization

Explainable AI (XAI) in Video Summarization combines two advanced fields: video
summarization, which condenses long videos into concise, meaningful summaries, and XAI,
which makes AI decisions interpretable to humans. The goal of XAI is to improve
transparency, accountability, and trust by explaining how decisions are made and
identifying the factors influencing them. For instance, in surveillance, a system analyzes
long video footage to generate summaries of important events, like detecting intrusions or
identifying vehicles, while offering explanations for its choices.

Workflow in Video Summarization with XAI
1.Video Summarization Model: The AI identifies and extracts key segments of the video
based on motion, objects, or anomalies, creating a condensed summary focused on
relevant events.
2.Explainability Features:
Attention Maps: Heatmaps overlay video frames to show areas of focus, such as
detected motion or objects.
Textual Explanations: Each clip is explained, e.g., "This segment was chosen due to a

moving vehicle entering a restricted area."
3.Human-in-the-Loop: Users can adjust summarization criteria, such as prioritizing
vehicles over people, allowing the system to refine its outputs.

How XAI Enhances Video Summarization
XAI ensures the video summarization process is interpretable:
Feature Importance: Highlights visual, audio, or contextual features that influence clip
selection, such as motion or object detection.
Attention Maps: Shows which areas of the video the model focused on using
techniques like Grad-CAM.
Textual Explanations: Provides readable reasons for clip selection, e.g., "Selected due
to high motion and a detected face."
Human-in-the-Loop: Users can interact with the system to adjust criteria and customize
summaries.