Paper Info
Title
Crime prediction and mapping based on real time video analysis.
Abstract
This paper presents a three phase approach to crime prediction based on video analysis, neuro-fuzzy inference and density mapping. In the first phase, crime indicator concepts are modeled and used in building classifiers for crime indicator events. Both indicator concept modeling and indicator event classification are performed using Generalized Maximum Clique Problem (GMCP) method. In the second phase, a neuro-fuzzy inference system modeled from training data is used to make predictions about classified crime indicator events obtained from the first phase. Finally in the third phase, kernel density estimation (KDE) is used to fit a spatial probability density function to the predicted crime indicator events across the study area. The major advantages of this method include the potential to predict crime in real time due to the use of video based events, the ability to generate fuzzy rules from data, the ability to optimize fuzzy rule-base by learning and the ability of weighting different crime variables. The proposed framework has prospects for developing a police field decision support system. The feasibility of the framework has been tested in a simulated experiment using sampled clips from violent scene detection (VSD) 2014, Hollywood Human Action (HOHA) and HMDB datasets and the results are quite promising for real life implementation.
Year
DOI
Venue
2018
10.3233/AIS-180476
JOURNAL OF AMBIENT INTELLIGENCE AND SMART ENVIRONMENTS
Keywords
Field
DocType
Video analysis,smart surveillance,crime prediction,crime mapping,neuro-fuzzy inference
Computer science,Human–computer interaction
Journal
Volume
Issue
ISSN
10
2
1876-1364
Citations 
PageRank 
References 
0
0.34
13
Authors
4
Name
Order
Citations
PageRank
Mohammed Nurudeen110.70
Beiji Zou223141.61
chengzhang zhu3153.91
Rongchang Zhao4304.63