Analysing social interactions through a wearable camera: a first-person point of view
| Autor: | Felicioni, Simone |
|---|---|
| Přispěvatelé: | Costante, Gabriele, Dimiccoli, Mariella |
| Rok vydání: | 2020 |
| Předmět: | |
| Zdroj: | Digital.CSIC. Repositorio Institucional del CSIC instname |
| Popis: | Nowadays social interaction analysis in egocentric vision is an increasingly popular topic due to its wide range of potential applications in domains including social robotics, social care and assistive technology. This Thesis aims to develop an automatic social interaction classification approach for egocentric videos captured by a wearable camera. The proposed method consists of four major steps. The first step is the extraction of social cues in videos. The second step is the construction of a graph based on the extracted features, where people in the scene can be seen as nodes and relations between them as edges. Then, a Relational Graph Convolutional Network is used on the graph to extract feature at the frame level. The third step is based on a gated recurrent unit to extract context information from the sequence of frames. The last step is the classification of the social interactions seen in the scene. Performances of this approach are evaluated with the Fathi et al. [1] first-person social interaction dataset. The current state of the art had never tried to deal with this problem using graph based neural networks, but experimental results show that this method is a valid mean to classify social interactions in egocentric video. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|