Related Experiment Videos
Panoramic appearance-based recognition of video contents using matching graphs
Chu-Song Chen1, Wen-Teng Hsieh, Jiun-Hung Chen
1Institute of Information Science, Academia Sinica, Taipei, Taiwan, ROC. song@iis.sinica.edu.tw
Summary
This study introduces a novel video content recognition method using panoramic images. The approach effectively identifies video content by treating recognition as a shortest-path problem solved with dynamic programming.
Area of Science:
- Computer Vision
- Robotics
- Artificial Intelligence
Background:
- Environmental representation is crucial for visual recognition tasks.
- Panoramas offer a compact, omni-directional representation of scenes.
- Existing video recognition methods often rely on single frames.
Purpose of the Study:
- To propose a general scheme for video content recognition.
- To leverage panoramic image databases for enhanced recognition.
- To utilize sequences of video frames for improved visual understanding.
Main Methods:
- Formulating video recognition as a shortest-path searching problem.
- Employing dynamic programming techniques to solve the shortest-path problem.
- Utilizing a database of panoramas for environmental context.
Main Results:
- Experimental results demonstrate the effectiveness of the proposed method.
- The system successfully recognizes video content using panoramic data.
- Sequential frame analysis improves recognition accuracy.
Conclusions:
- The proposed scheme provides an effective approach for video content recognition.
- Using panoramas and sequential frames enhances environmental recognition.
- Dynamic programming offers a viable solution for this recognition task.