Two-person interaction recognition using a two-step sequential pattern classification.
Saved in:
| Title: | Two-person interaction recognition using a two-step sequential pattern classification. |
|---|---|
| Authors: | Nikzad, Saman1 (AUTHOR) s_nikzad@sut.ac.ir, Ebrahimi, Afshin1 (AUTHOR) aebrahimi@sut.ac.ir |
| Source: | Multimedia Tools & Applications. Apr2025, Vol. 84 Issue 12, p10465-10487. 23p. |
| Subjects: | Sequential pattern mining, Information storage & retrieval systems, Artificial intelligence, Computer vision, Image processing |
| Abstract: | As image processing techniques and devices advance, the real-time applications of computer vision such as human action/interaction recognition and video content analysis become more attractive. However, the methods proposed in the state-of-the-art studies are still far from representing real-time and all-inclusive classifiers because of the image and video analysis complexity. This work presents a new approach based on key-poses of frame silhouettes for human interaction recognition. We use an inner-distance-based shape descriptor which gives a perfect description of the shape due to its ability to collect data from the whole shape. The core idea is to develop a two-step classifier based on a sequential pattern mining classifier. So, we extract the Bilateral Silhouette shape for the persons and describe it based on the inner-distance feature to compare each frame with a pre-defined dictionary of key-poses. The classification process is performed in frame and sequence layers. Accurate and efficient, the sequential pattern mining approach provides an appealing solution to the problem of sequence classification, giving comparable or even better results than standard classifiers. We evaluated the recognition performance of the system using video sequences of SBU human interaction dataset and the UT-interaction dataset as two well-known interaction datasets and the results are considered acceptable (95.25% in SBU and 90.5% in UT databases, respectively), outperforming most state-of-the-art results. These recognition rates are calculated after we have tested different parameters which can affect the results. Both datasets include multiple interaction classes performed by different actors, which helps us develop an all-inclusive method based on the datasets. The proposed method can be optimized to be used in some real world applications such as abnormal activity recognition in crowded places, auxiliary surveillance system, human-computer interaction, etc. [ABSTRACT FROM AUTHOR] |
| Copyright of Multimedia Tools & Applications is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
|
Full text is not displayed to guests.
Login for full access.
|
|
Be the first to leave a comment!
