Scholarlink

<p>A data-driven Intelligent Transportation Systems (ITS) framework for advanced traffic video analysis that automatically detects, identifies and interprets critical traffic scenarios. Meanwhile, foster new insights and develop deep learning models to overcome the challenges in the transportation system and urge sustainable solutions for real-world problems. The proposed system covers advanced deep learning-based video processing, real-time accident detection, Explainable Artificial Intelligence (XAI), and visualization tools through enhanced Video Question Answering (VideoQA) frameworks. This pipeline consists of accident detection analysis, real-time traffic monitoring, and management, including counting passengers and vehicle flow, etc., which are applied on custom datasets. To foster an enhanced Video Question Answering (VideoQA) system that generates textual answers, both frame and a segment of video clips based on user queries. The hybrid mechanism of Timesformer for video feature extraction and a more meticulous text encoder called Sentence Transformer was incorporated to develop a VideoQA pipeline. Additionally, an explainable Artificial Intelligence (XAI) focuses on the hidden areas from the response of the best frame based on the user queries, and the pipeline shows superior performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced multi-modal Video Question Answering framework in Intelligent Transportation Systems with explainable AI

  • 期刊论文

  • 发表日期：

    2026年6月23日

查看全文
  • P. J Jeshmol,
  • Binsu C. Kovoor

摘要

A data-driven Intelligent Transportation Systems (ITS) framework for advanced traffic video analysis that automatically detects, identifies and interprets critical traffic scenarios. Meanwhile, foster new insights and develop deep learning models to overcome the challenges in the transportation system and urge sustainable solutions for real-world problems. The proposed system covers advanced deep learning-based video processing, real-time accident detection, Explainable Artificial Intelligence (XAI), and visualization tools through enhanced Video Question Answering (VideoQA) frameworks. This pipeline consists of accident detection analysis, real-time traffic monitoring, and management, including counting passengers and vehicle flow, etc., which are applied on custom datasets. To foster an enhanced Video Question Answering (VideoQA) system that generates textual answers, both frame and a segment of video clips based on user queries. The hybrid mechanism of Timesformer for video feature extraction and a more meticulous text encoder called Sentence Transformer was incorporated to develop a VideoQA pipeline. Additionally, an explainable Artificial Intelligence (XAI) focuses on the hidden areas from the response of the best frame based on the user queries, and the pipeline shows superior performance.