UTAR Institutional Repository

Violent scene detection in videos

Yew, Kynn Man (2019) Violent scene detection in videos. Final Year Project, UTAR.

Download (3250Kb) | Preview


    Long term exposed to violent content may cause harm to an individual, especially younger children. Nowadays, video sharing and streaming websites are becoming more and more widespread which makes exposure to unwanted violent content much more frequent and inevitable. While there exist many types of violent content, encountering content with physical violence seems to be more common compared other types of violence. Unlike the other types of violent content, physical violence can usually be identified by using visual cues and often associated with certain actions. Recent Convolutional Neural Networks (CNN) has shown great success in visual tasks such as image recognition and classification tasks. Furthering the success, Convolutional Neural Networks extended to video data and has shown that CNN can effectively extract and learn important features for complex tasks such as human action recognition. In this project, a desktop application for violent scene detection and localisation is built using multimodality deep learning architecture for violent scene detection. Besides that, this application also provides users with the interface to view and filter detected violent scenes. To examine the generalisation capability of this application on different video types, this application is tested on Hollywood movies and web-based videos. Based on testing results, despite less optimal training process, this application is able to detected violent scene in Hollywood movies quite well, MAP2104 in Hollywood movies is 0.56. The performance on web-based videos on the other hand is poorer with MAP2014 of 0.43.

    Item Type: Final Year Project / Dissertation / Thesis (Final Year Project)
    Subjects: Q Science > Q Science (General)
    Divisions: Faculty of Information and Communication Technology > Bachelor of Computer Science (Honours)
    Depositing User: ML Main Library
    Date Deposited: 20 Aug 2019 12:15
    Last Modified: 20 Aug 2019 12:15
    URI: http://eprints.utar.edu.my/id/eprint/3494

    Actions (login required)

    View Item