Journal article
Body detection in spectator crowd images using partial heads
Image and Video Technology, Vol.11854, pp.65-77
2019
Abstract
In spectator crowd images, the high number of people, small size and occlusion of body parts, make the body detection task challenging. Due to the similarity in facial features of different people, the variance in head features is less compared to the variation in the body features. Similarly, the visibility of the head in a crowd is more, compared to the visibility of the body. Therefore, the detection of only the head is more successful than the detection of the full body. We show that there exists a relation between head size and location, and the body size and location in the image. Therefore, head size and location can be leveraged to detect full bodies. This paper suggests that due to lack of visibility, more variance in body features, and lack of available training data of occluded bodies, full bodies should not be detected directly in occluded scenes. The proposed strategy is to detect full bodies using information extracted from head detection. Additionally, body detection technique should not be affected by the level of occlusion. Therefore, we propose to use only color matching for body detection. It does not require any explicit training data like CNN based body detection. To evaluate the effectiveness of this strategy, experiments are performed using the S-HOCK spectator crowd dataset. Using partial ground truth head information as the input, full bodies in a dense crowd is detected. Experimental results show that our technique using only head detection and color matching can detect occluded full bodies in a spectator crowd successfully.
Details
- Title
- Body detection in spectator crowd images using partial heads
- Authors/Creators
- Y. Jan (Author/Creator) - Murdoch UniversityF. Sohel (Author/Creator) - Murdoch UniversityM.F. Shiratuddin (Author/Creator) - Murdoch UniversityK.W. Wong (Author/Creator) - Murdoch University
- Publication Details
- Image and Video Technology, Vol.11854, pp.65-77
- Publisher
- Springer Verlag
- Identifiers
- 991005542896407891
- Copyright
- © 2019 Springer Nature Switzerland AG
- Murdoch Affiliation
- Information Technology, Mathematics and Statistics
- Language
- English
- Resource Type
- Journal article
- Additional Information
- Conference paper from Pacific-Rim Symposium on Image and Video Technology (PSIVT) 2019. Sydney, NSW 18 - 22 November 2019
UN Sustainable Development Goals (SDGs)
This output has contributed to the advancement of the following goals:
Source: InCites
Metrics
108 Record Views