A Deep Learning Model for Anomaly Detection in Video Using Object Segmentation and Localization
13th International Conference on Electrical and Electronics Engineering, ICEEE 2026, Antalya, Türkiye, 27 - 29 Nisan 2026, ss.577-583, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/iceee69936.2026.11598224
- Basıldığı Şehir: Antalya
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.577-583
- Anahtar Kelimeler: Anomaly detection, localization, object segmentation, video analytics
- Gazi Üniversitesi Adresli: Evet
Özet
Video anomaly detection is an important problem for intelligent surveillance, public safety, and smart environment monitoring. Traditional approaches often focus on classification, thereby failing to detect anomalous events in real time. This paper proposes a novel multimodal anomalydetection model that integrates text-based anomaly interpretation, named entity recognition (NER), open-world object detection, segmentation, and depth-aware localization. The proposed system combines a vision-based anomaly description model, NER-based keyword extraction, the segment anything model (SAM) for segmentation, and Video Depth Anything for monocular depth estimation. Sensor-based depth measurements from Intel RealSense cameras are used in the experiments. Experiments on real-world videos show that the proposed pipeline provides semantically grounded and spatially localized anomaly analysis, enabling both pixel-level and objectlevel reasoning.