Case study
AI-Powered Unattended Object Detection System Delivering 90%+ Detection Accuracy for Public Safety and Security
MoreYeahs built a camera-based intelligent surveillance system that detects, classifies, and tracks objects and people in real time to flag unattended items automatically.
- Industry
- Public Sector & Transit Security
- Engagement
- AI-Powered Unattended Object Detection
- Focus
- Real-Time Security Monitoring for Public Infrastructure
Results at a glance
- 0190%+ unattended object detection accuracy achieved in test scenarios.
- 0285%+ owner-object association accuracy achieved in test scenarios.
- 03Alerts triggered within 5 seconds of a dwell-time threshold breach.
01The client
Customer
Airports, railway stations, bus terminals, and metro stations are exposed to security risks from unattended or unidentified objects such as bags, suitcases, and packages, which may indicate theft risk or a potential hazard.
MoreYeahs built an AI-powered unattended object detection system for organizations securing this kind of public infrastructure, giving control room operators automated visibility that manual CCTV observation alone cannot provide.
02Challenge
Manual monitoring left security teams reactive rather than proactive:
- 01Resource-Intensive Manual Observation
Monitoring relied heavily on operators watching CCTV feeds continuously, which is resource-intensive and prone to human oversight.
- 02Delayed Identification
Unattended objects often went unnoticed for extended periods, reducing response time.
- 03No Owner Correlation
There was no automated way to link an object to the person who left it or to track how long it had been unattended.
- 04Scale Across Busy Facilities
High-traffic transit hubs generate more visual data than operators can reliably track.
03Solution
Solution
MoreYeahs built a camera-based intelligent surveillance system that detects, classifies, and tracks objects and people in real time to flag unattended items automatically.
Computer vision models detect and classify bags, suitcases, and packages in the camera feed.
Multi-object tracking correlates each object with its associated person.
Objects that remain stationary while their owner is absent beyond a configurable threshold are flagged as unattended.
Flagged events generate alerts with location and a dashboard highlight, plus a snapshot and video clip for security personnel.
04How we delivered
The system continuously tracks the relationship between people and objects to decide when an item is genuinely unattended.
- Step 1Continuous Tracking
Live CCTV video is ingested and analyzed, with multi-object tracking assigning persistent IDs to people and objects.
- Step 2Trajectory-Based Association
Objects are associated with the nearest person based on trajectory and proximity.
- Step 3Dwell-Time Escalation
When an object stays stationary and its owner moves away past the configured threshold, an alert is generated with camera ID and zone coordinates.
- Step 4Exception Handling
Temporary owner occlusion applies a short grace period, and objects moved by another person trigger a dwell-time reset rather than a false alert.
05Technology
The solution runs on high-resolution CCTV infrastructure paired with object detection and multi-object tracking models.
- Object Detection & Tracking Models
- Computer vision models for object classification and persistent tracking
- Edge/Server Inference
- Infrastructure sized for continuous multi-camera analysis
- Dwell-Time Rule Engine
- Configurable thresholds for unattended-object classification
- Alerting, Dashboard & Storage
- Video management system integration with an event storage database
06Results
In validation testing, the system met its target benchmarks for detection accuracy and alert speed.
- 90%+ unattended object detection accuracy…
90%+ unattended object detection accuracy achieved in test scenarios.
- 85%+ owner-object association accuracy achieved…
85%+ owner-object association accuracy achieved in test scenarios.
- Alerts triggered within 5 seconds…
Alerts triggered within 5 seconds of a dwell-time threshold breach.
- System handled moderate crowd density…
System handled moderate crowd density without major degradation in detection quality.
07Business impact
Automated detection gives security teams at public infrastructure sites earlier, more reliable awareness of potential threats.
Automated, location-tagged alerts cut the time to identify and inspect unattended items.
Continuous AI monitoring reduces reliance on constant manual CCTV observation.
The system extends consistent monitoring across multiple zones of a busy transit facility simultaneously.
Automatic snapshots and video clips support rapid verification and post-incident review.
More projects like this
All case studiesFacing a similar challenge?
Tell us what you need. A technical lead will reply within one business day with how we would approach it.
- Reply within one business day
- Free 30-minute call
- NDA before you share data