- Version
- Download 4
- File Size 621.93 KB
- File Count 1
- Create Date 21/04/2026
- Last Updated 21/04/2026
AI-POWERED CRIME DETECTION SYSTEM
Deeksha Ahirwar
Student
Department of Computer Science & Engineering
University Institute of Technology
Barkatullah University , Bhopal
deekshaa0804@gmail.com
Pankaj Shah
Student
Department of Computer Science & Engineering
University Institute of Technology
Barkatullah University , Bhopal
pankaj92shah@gmail.com
Anushree Bhargava
Student
Department of Computer Science & Engineering
University Institute of Technology
Barkatullah University , Bhopal
anushreebhargava2004@gmail.com
Harshita Tripathi
Student
Department of Computer Science & Engineering
University Institute of Technology
Barkatullah University , Bhopal
tripathiharshita393@gmail.com
Dr. Kavita Chourasiya
Co-guide
Department of Computer Science & Engineering University Institute of Technology
Barkatulla University , Bhopal
chourasiakavita635@gmail.com
Dr. Kamini Maheshwari
Guide
Department of Computer Science & Engineering
University Institute of Technology
Barkatullah University , Bhopal
kaminimaheshwari@gmail.com
Dr. Divakar singh
H.O.D
Department of Computer Science & Engineering
University Institute of Technology
Barkatullah University , Bhopal
divakarsingh@gmail.com
Abstract
Loitering, defined as the prolonged presence of an individual or group in a particular area without a clear purpose, has been identified as a significant indicator of potential criminal or suspicious activity in public and private spaces. Traditional security systems rely heavily on human surveillance, which is prone to fatigue, bias, and error, especially in environments requiring continuous monitoring such as malls, parking lots, transportation terminals, and critical infrastructure zones. To address these limitations, this research presents the design and implementation of an AI-based loitering detection system that leverages deep learning and computer vision techniques to identify, track, and analyze human behavior in real-time. The proposed system integrates the YOLOv8 (You Only Look Once, version 8) object detection model with a custom tracking and time-based behavioral analysis module, enabling precise recognition of individuals who exhibit loitering behavior within a designated field of view.
The system architecture is composed of three core modules: detection, tracking, and behavior analysis. The detection module employs YOLOv8, a state-of-the-art real-time object detection framework known for its high accuracy and computational efficiency. By fine-tuning the YOLOv8n model with datasets representing various surveillance scenarios, the system accurately identifies human subjects while minimizing false positives from non-human objects. Once individuals are detected, the tracking module maintains a persistent identity across frames using an object-tracking algorithm. This ensures that movement patterns are consistently monitored, even when partial occlusion or overlapping occurs. The final module, behavioral analysis, interprets the spatiotemporal data generated by tracking to determine whether an individual has remained within a specific region for longer than a predefined threshold, signifying loitering behavior.
