Suspicious Behavior Detection with Temporal Feature Extraction and Time-Series Classification for Shoplifting Crime...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Rising global crime rates necessitate advanced automated detection solutions.
• Current computer vision methods for crime detection often rely on spatial features from pixel data.
• Shoplifting prevention requires effective automated suspicious behavior detection.

Purpose of the Study:

• To develop a novel approach for detecting suspicious behavior to prevent shoplifting.
• To improve the efficiency and accuracy of automated crime detection systems.
• To overcome limitations of existing spatial feature-based methods.

Main Methods:

• Utilized YOLOv5 object detection and Deep Sort for human tracking in videos.
• Extracted temporal features from bounding box coordinates for time-series classification.
• Benchmarked against the state-of-the-art Robust Temporal Feature Magnitude (RTFM) method using Inflated 3D ConvNet (I3D).

Main Results:

• Achieved an 8.45-fold increase in detection inference speed compared to RTFM.
• Attained an F1 score of 92%, outperforming RTFM by 3%.
• Demonstrated effectiveness without requiring expensive data augmentation or image feature extraction.

Conclusions:

• The proposed temporal feature-based method offers a significant advancement in automated suspicious behavior detection.
• This approach provides a faster and more accurate solution for shoplifting prevention.
• The method's efficiency makes it a practical tool for real-world crime detection applications.