PROPOUNDING FIRST ARTIFICIAL INTELLIGENCE APPROACH FOR PREDICTING ROBBERY BEHAVIOUR POTENTIAL IN AN INDOOR SECURITY CAMERA

Abstract

For video surveillance systems to stop 
incidents and safeguard assets, crime 
prediction is necessary. In this regard, our 
paper presents the first artificial intelligence 
method for indoor camera Robbery Behavior 
Potential (RBP) identification and prediction. 
Three detection modules—the head cover, 
crowd, and loitering detection modules—are 
the foundation of our approach, which enables 
us to take prompt action and stop robberies. 
Using our collected, manually annotated 
dataset, the YOLOV5 model is retrained for 
the first two modules. Additionally, we 
provide a brand-new definition for the 
DeepSORT-based loitering detection module. 
A fuzzy inference computer converts expert 
information into rules before making a final 
determination about the likelihood of a heist. 
This is difficult because the thief uses a 
different technique, the security camera is 
angled differently, and the video pictures have 
poor quality.Using actual video surveillance 
footage, we successfully completed our 
experiment and obtained an F1-score of 0.537. 
Thus, we develop threshold value for RBP to 
assess video pictures as a robbery detection 
issue and compare experimentally with other 
similar research. Assuming this, the 
experimental findings clearly reflect an F1-
score of 0.607, indicating that the suggested 
technique outperforms other robbery detection 
methods in terms of identifying robberies. We 
firmly think that by anticipating and averting 
robbery incidents, the suggested method's 
implementation might reduce the harm caused 
by robberies at a security camera control 
center. However, a human operator's 
situational awareness improves and additional 
cameras may be controlled.