University of Twente Student Theses


Exploring Multimodal Data for Crime Recognition

Poozhiyil, Aditya (2023) Exploring Multimodal Data for Crime Recognition.

[img] PDF
Abstract:With the generation of diverse forms of data being produced exponentially by various forms of devices, researchers have explored exploiting the inherent characteristics of the modality to recognize human actions. The most widely generated data, RGB(color) primarily contributes to the spatial information but lacks the temporal attributes. Conversely, the skeleton modality emphasizes the temporal aspect of the human joints but lacks spatial features. Both these modalities present features that can mutually complement each other. In the context of crime recognition, earlier research focused on capturing and learning temporal patterns by exploring different forms of Transformer architectures with skeleton trajectories. This study extends the work by investigating the fusion of visual context(RGB) with the skeleton to leverage the spatial and temporal dynamics of both modalities. The dataset used for this study is the HR-Crime, containing 13 human-related crime categories captured through surveillance cameras. Our experiments show the fusion of both modalities shows improvement compared to the baseline. In addition, we discuss the limitations of our approach and possible ways to tackle them.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science MSc (60300)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page