Propounding First Artificial Intelligence Approach For Predicting Robbery Behavior Potential In An Indoor Security Camera
Keywords:
Robbery Behavior Prediction, AI Surveillance, Crime Prevention, YOLOv5, Deep SORT, Fuzzy Logic, Anomaly Detection, Security Cameras, Computer Vision, Behavior Analysis, Indoor Security, Public Safety AI.Abstract
The project on "Crime Prediction in Video-
Surveillance Systems" presents a forwardthinking
AI-driven surveillance solution for
enhancing public safety by predicting and
detecting Robbery Behavior Potential (RBP)
in indoor environments through video
analysis. The system integrates three
specialized detection modules - head cover
identification, crowd density analysis, and
loitering behavior tracking - each targeting
early suspicious activity indicators. For head
cover and crowd detection, we retrained the
YOLOv5 object detection model using a
custom annotated dataset, while introducing a
novel Deep SORT tracking algorithm
implementation for precise loitering behavior
analysis. These components feed into a fuzzy
inference engine that applies expert-defined
rules to assess robbery threats, addressing
significant challenges including behavioral
variability, diverse camera angles, and
typically low-resolution footage. During realworld
testing on surveillance videos, the
system achieved an initial F1-score of 0.537,
which improved to 0.607 when evaluating
RBP against a defined robbery detection
threshold - outperforming existing methods.
This demonstrates the system's potential to not
only detect but proactively prevent robberies,
thereby reducing losses and significantly
enhancing situational awareness for security
operators in control centers.
