Integrated AI Pipeline for Real-Time Anomaly Detection and Depth Analysis
- Project
- 22006 SINTRA
- Type
- New service
- Description
Developed and validated a multi-model pipeline combining HolmesVAU (LLM-based anomaly detection), NER, and YOLO World for dynamic object recognition in video streams. Implemented depth estimation using Video Depth Anything and cross-validated with Intel RealSense sensor data. Real-world data collection at Esenboğa Airport and privacy-preserving face blurring were successfully completed, enabling advanced situational awareness for security and surveillance.
- Contact
- aylin.yorulmaz@kocsistem.com.tr
- aylin.yorulmaz@kocsistem.com.tr
- Research area(s)
- Video Anomaly Detection, Depth Estimation, Multimodal AI, Privacy-Preserving Analytics
- Technical features
Developed and validated a multi-model pipeline combining HolmesVAU (LLM-based anomaly detection), NER, and YOLO World for dynamic object recognition in video streams. Implemented depth estimation using Video Depth Anything and cross-validated with Intel RealSense sensor data. Real-world data collection at Esenboğa Airport and privacy-preserving face blurring were successfully completed, enabling advanced situational awareness for security and surveillance.
- Integration constraints
TBD
- Targeted customer(s)
TBD
- Conditions for reuse
TBD
- Confidentiality
- Public
- Publication date
- 30-04-2026
- Involved partners
- TAV Technologies (TUR)
- KoçSistem (TUR)