GIST Develops 'VSCDNet' for Automatic Object Change Detection Across Different Filming Times
Analyzing Entire Videos to Detect Disappearance and Movement Changes
The research team led by Professor Uiwhan Kim from the Department of AI Convergence at Gwangju Institute of Science and Technology (GIST) has announced the development of an artificial intelligence (AI) model called 'VSCDNet.' This model can automatically detect actual object changes by comparing videos of the same space taken at different times and along different paths.
(From left) Professor Eehwan Kim of the AI Convergence Department, Jiae Yoon, integrated master's and doctoral student.
View original imageUnlike previous methods which compared individual photos, this model analyzes entire videos to identify changes such as the appearance, disappearance, and movement of objects.
For autonomous robots to move independently in indoor spaces over extended periods, they must continuously recognize changes in their surroundings. However, traditional change detection technologies mainly compare images taken from similar locations and at similar times, which limits their accuracy in detecting changes when videos are filmed along different paths.
To overcome these limitations, the research team focused on the overall flow of the video instead of comparing individual images. The developed AI model, 'VSCDNet,' compares reference videos taken in the past with current videos filmed in the same space. It identifies corresponding scenes throughout the videos and precisely detects only the areas where actual object changes have occurred.
Based on this, the model generates a 'change mask' that visually highlights the areas where changes have taken place, presenting the final regions of change. In other words, it can automatically detect actual object changes, such as a laptop disappearing or the position of an object shifting.
To systematically verify the model's performance, the research team constructed a large-scale dataset that includes both virtual spaces and real indoor environment data.
The dataset consists of a total of 1,090 videos (over 1.13 million frames). Experimental results showed that 'VSCDNet' demonstrated the best performance compared to existing change detection techniques on both the virtual space dataset and the real indoor environment dataset. It maintained stable detection performance under various conditions, such as different video lengths, image qualities, and numbers of changed objects.
Additionally, in experiments using actual mobile robots, the system automatically detected situations where doors opened or objects disappeared in videos recorded while robots moved along alternative paths. It also demonstrated the ability to remember and learn newly appearing objects.
This research is significant in that it enables the stable detection of only actual object changes, even in videos of the same space taken at different times and along different routes. In particular, its consistent performance across diverse environmental conditions enhances its potential for real-world application. Through experiments with real robots, the model's capabilities for monitoring environmental changes and learning new objects were also verified. It is expected to be applied in various fields, including indoor patrol robots, smart security surveillance, facility management, and IoT-based smart indoor systems.
Professor Uiwhan Kim stated, "VSCDNet is an AI model that goes beyond recognizing the current scene to autonomously identifying what has changed compared to the past. Because it can compare videos filmed along different routes without requiring separate location information or spatial maps, we expect it to be utilized in areas such as indoor patrol robots, smart security surveillance, and facility management."
Hot Picks Today
"Two Spend 500,000 Won in a Flash, Yet Wait Two Hours?" ... 'Dreamland' Shunned by Kids and Foreign Tourists
- The Era of Getting 200 Million Won After 30 Years Is Over... A Japanese Bank's "Perpetual Department Head" Who Just Clocked In Cries Out
- Spain Hit by Its Worst Tragedy... Lee Kangin, Moreno and Real Madrid Join Soccer World in Mourning
- They Said, “Make It Your Home for Life” - Then Suddenly Kicked 200 Passengers Off... What Happened to the Around-the-World Cruise?
- "130 Rice Bowls Sold Out in a Day"... 'Unfiltered Face' Adds 8 Million Won to Night Market Sales: China's Bungeunwoo
This research, supervised by Professor Uiwhan Kim from the Department of AI Convergence and conducted by integrated master's and doctoral program student Jiae Yoon, was supported by the Excellent Young Researcher Support Program of the Ministry of Science and ICT and the National Research Foundation of Korea, the GIST-InnoCORE Project, the Self-Directed Visual Intelligence Technology Development Project of the Institute for Information & Communications Technology Planning & Evaluation, and the Global TOP Strategic Research Group Support Project of the National Research Council of Science & Technology. The results will be presented at 'ICML 2026 (International Conference on Machine Learning),' the world's top conference in the AI and machine learning field, to be held at COEX Convention Center in Seoul from July 6 to 11. Prior to this, the paper was published on the preprint site 'arXiv' on May 20.
© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.