AI Dev 26 x SF | Thierry Damiba: Edge to Cloud Video Anomaly Detection

2026-05-22 · Source: DeepLearningAI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Internet of Things (IoT) & Connected Devices, Cloud Computing & IT Infrastructure · Depth: Intermediate, long

Summary

A system for edge-to-cloud video anomaly detection utilizes vector search on edge devices to significantly reduce cloud compute needs. This architecture uses Quadrant Edge, running EfficientNet B0 on NVIDIA Jetson, to generate video embeddings and detect anomalies based on dissimilarity in a vector index. Events exceeding a threshold are then sent to 12 Labs in the cloud for detailed analysis, with Quadrant Cloud continuously syncing baseline definitions back to the edge. The system achieves 94% recall on 30-second videos, resulting in approximately two false positives per hour, and reduces bandwidth by 90%. Demonstrated through the "Sentinel" application, it monitors 12 cameras across three zones, providing AI-generated alerts, textual filtering, and a vector space visualization of clips. The approach focuses on identifying deviations from a learned "normal" baseline, rather than immediate classification, and applies vector search principles to transform video clips into numerical representations for efficient similarity and dissimilarity detection.

Key takeaway

For MLOps Engineers deploying real-time video analytics, this edge-to-cloud vector search architecture offers a compelling solution to manage data volume and cloud costs. You should consider implementing edge-based anomaly detection with baseline synchronization to offload 90% of video processing from the cloud. This approach ensures high recall for critical events while significantly reducing your operational expenses and bandwidth usage.

Key insights

Edge-based vector search for video anomaly detection significantly reduces cloud compute by identifying deviations from a learned baseline.

Principles

• Anomaly detection is finding dissimilarity in vector space.
• Prioritize recall to avoid missing true anomalies.
• Baselines must dynamically sync between edge and cloud.

Method

Generate video embeddings on edge devices, detect anomalies via dual shard KNN dissimilarity, escalate high-threshold events to cloud, and sync baseline vector index from cloud to edge.

In practice

• Use EfficientNet B0 on NVIDIA Jetson for edge embeddings.
• Implement semantic search for complex queries like "outfit for my daughter's wedding who is a little snobby".
• Visualize vector space to analyze clip groupings and anomaly distances.

Topics

• Edge AI
• Video Anomaly Detection
• Vector Search
• MLOps
• NVIDIA Jetson
• Semantic Search

Best for: AI Architect, Computer Vision Engineer, CTO, AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• AI-ready infrastructure: The foundation for scalable AI workloads — AI success depends on robust infrastructure capable of rapid data movement, predictable scaling, and reliable operations, rather than solely on model strength.
• EdgeZSAD: Practical Zero-Shot Anomaly Detection on Edge Devices — EdgeZSAD enables high-performance zero-shot anomaly detection on edge devices using a compact TinyViT backbone and specialized readout.
• Getting a transit-fraud detector into production by fixing the features, not the model — Effective fraud detection prioritizes feature engineering over model complexity, especially for evolving adversarial patterns.
• Building a High-Performance Vector Search Engine from Scratch in 2026 — Building a custom vector search engine offers maximum privacy and zero latency overhead for AI applications.
• Token-Efficient Change Detection in LLM APIs — B3IT enables cost-effective, black-box LLM change detection using specific "Border Inputs" that yield multiple top output tokens.
• New Relic and Agentic DevOps with Nic Benders — Observability is evolving from passive monitoring to AI-driven active intelligence, automating problem detection and resolution.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by DeepLearningAI.