Abnormal High Density Crowds

Citation Author(s):
Sanjeeb
Tiwary
Submitted by:
sanjeeb tiwary
Last updated:
Fri, 11/10/2023 - 10:52
DOI:
10.21227/m4vb-p620
Data Format:
License:
0
0 ratings - Please login to submit your rating.

Abstract 

Data Preprocessing:

All video footage in this dataset has undergone comprehensive preprocessing techniques, including frame extraction, anomaly identification, cropping of frames to emphasize crowded scenes, and compression to reduce storage space.

Overview:

This dataset encompasses four distinct scene incidents, each meticulously detailed below:

 

1. Times Square Chaos:

Video Source: Times Square Chaos from Three Angles

 

Viewpoints: Three different perspectives.

View_1: Angled shot, abnormal behavior starts at 00:00:12. Dimensions: 1280x720.

View_2: Closeup eye-level shot, abnormal behavior starts at 00:00:53. Dimensions: 1280x720.

View_3: Angled shot with cropped dimensions (580x720), abnormal behavior starts at 00:01:39.

2. Las Vegas Mass Shooting:

Video Source: Las Vegas Mass Shooting CCTV Video

 

Training & Testing Frames: Divided into normal and abnormal instances at different angles and time frames.

Train: Wide angled shot, normal behavior between 00:11:17 and 00:16:05. Dimensions: 992x468.

Test_1: Same angle as training footage, abnormal behavior between 00:16:06 and 00:17:17. Dimensions: 992x468.

Test_2, Test_3, Test_4: Varied angles and grayscale for abnormal behavior instances. Dimensions: 1280x720.

3. Love Parade Disaster:

Video Source: Love Parade Disaster

 

Training & Testing Frames: Split into normal and abnormal sequences at a specific angle and time frame.

Train_1, Train_2: Angled shots with normal behavior within defined time frames. Dimensions: 1280x720.

Test: Abnormal behavior between 00:10:48 and 00:11:03. Dimensions: 1280x720. Frame-level labeling stored in XML format.

4. Juventus Fan Panic in Italy:

Video Sources: Video 1, Video 2

 

Viewpoints: Divided into two distinct viewpoints comprising abnormal behaviour instances.

View_1: Wide, close, eye-level shot with abnormal behavior between 00:00:00 and 00:00:28. Dimensions: 1280x720.

View_2: Wide angled shot, abnormal behavior between 00:00:00 and 00:00:28. Dimensions: 880x720.

This dataset provides a comprehensive collection of annotated abnormal behaviors captured in different real-life scenarios, crucial for anomaly detection research and analysis.

Instructions: 

All video footage in this dataset has undergone comprehensive preprocessing techniques, including frame extraction, anomaly identification, cropping of frames to emphasize crowded scenes, and compression to reduce storage space.