Feature and decision level fusion for action recognition

Abouelenien, Mohamed; Wan, Yiwen; Saudagar, A.

doi:10.1109/icccnt.2012.6477851

Cited by 2 publications

(2 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Fusion of multiple sources into an ensemble have been addressed by three main approaches in the literature: early, mid-level, and late [3]. Early fusion combines the information in the first possible level (so called signal level fusion in image processing), whereas late fusion combines the information as late as possible (decision level fusion) [4,5]. Mid-level fusion is an interesting compromise that combines the information in an intermediate abstraction level [6].…”

Section: Spatio-temporal Fusion For Learning Of Regions Of Interests mentioning

confidence: 99%

“…Some authors demonstrate [12,13] demonstrated the effectiveness of the decision level fusion strategies on object tracking, video segmentation, and video event detection. Feature-level fusion has gained much importance over the past few years, and various approaches have been introduced in the literature [5,14,15]. Most approaches combined the information of multiple modalities (sensors), while some methods used the complementary descriptors.…”

Section: Spatial-temporal Fusion Schemes Over Framesmentioning

confidence: 99%

See 1 more Smart Citation

Spatio-Temporal Fusion for Learning of Regions of Interests Over Multiple Video Streams

Khoshrou

Cardoso

Granger

et al. 2015

Advances in Visual Computing

View full text Add to dashboard Cite

of data captured over a network of cameras for various recognition tasks. In order to limit human labour and error, this paper presents a spatial-temporal fusion approach to accurately combine information from Region of Interest (RoI) batches captured in a multi-camera surveillance scenario. In this paper, feature-level and score-level approaches are proposed for spatial-temporal fusion of information to combine information over frames, in a framework based on ensembles of GMM-UBM (Universal Background Models). At the feature-level, features in a batch of multiple frames are combined and fed to the ensemble, whereas at the score-level the outcome of ensemble for individual frames are combined. Results indicate that feature-level fusion provides higher level of accuracy in a very efficient way. IntroductionVideo surveillance applications, such as activity recognition, are increasingly making use of multiple sensors and modalities. The fusion of multiple diverse sources of information is expected to benefit the system for the recognition of objects, persons, activities and events captured in an array of cameras.Networks of video cameras are commonly employed to monitor large areas for a variety of applications. A central issue in such networks is the tracking and recognition of individuals of interests across multiple cameras. These individuals must be recognized when leaving the Field of View (FoV) of one camera and re-identified when entering the FoV of another camera. Systems for video-to-video recognition are typically employed for person re-identification (PR). In a FoV, the appearance of an individual may be captured in reference RoIs and representative models may be learned from RoI trajectories. Then, the probe RoI may be matched against the reference model in either live (real-time monitoring) or archived (post-event analysis) [1]. In this paper, we address a PR system over wide network of cameras where no target individual enrolled to the system in advance.In such environments, where objects move and cross in the FoV of multiple cameras, it is likely to have multiple streams, recorded at different starting points with various lengths, for the same RoI of individuals (see Fig. 1a). The surveillance system must track that person across all cameras whose FoV overlap the person's path.

show abstract