Violence Detection in Campus Surveillance

DATASET OVERVIEW

Description

The dataset was specifically created to address the need for violence detection in surveillance systems. It

consists of self-recorded videos simulating different types of violent activities relevant to college

environments. The dataset is organized into four distinct classes:

Slap

Punch

Kick

Group Violence

Others - Over Crowding, Loitering, Assault, Abuse

Each video is labeled according to its corresponding class to facilitate supervised learning for violence

detection models.

DATA COLLECTION

Source: The videos were recorded in controlled settings, mimicking real-life scenarios on a college

campus.

Environment: Indoor and Outdoor

Equipment: Details about the cameras used - DSLR Camera

Actors: Simulated by volunteers following predefined scripts to ensure accuracy and variety.

Trimming: All videos were trimmed to ensure uniform duration across the dataset, maintaining

consistency for training and evaluation purposes.

DATASET STATISTICS

Class Distribution:

Slap: 30 videos

Punch: 25 videos

Kick: 30 videos

Group Violence: 50 videos

Over Crowding: 4 videos

Loitering: 4 videos

Assault: 10 videos

Abuse: 7 videos

Duration: 1-4 sec of video per class.

Resolution: 1920x1080

Video Format: mp4, avi, dav

FILES INCLUDED

folder with an excel of actor images, Name and Roll Number

folders of classes with videos