dusk idol
We have a marketing and security tool all in one (cloud solution, web app) for analyzing real-time camera feeds, including object tracking, scheduled notifications, face analysis, object counting, and restricted area monitoring, all built on Kubernetes in Azure. While the system is currently over-engineered, we need assistance to finalize it and make it production-ready.
Here are the necessary fixes and improvements:
Simplify the Architecture
API Service: Handle CRUD operations (cameras, ROI setup, etc.) in a single service.
Ingestion Service: Separate this service to handle video processing (detection, tracking, overlay) on the GPU.
Consumer Service: Handle metadata processing (bounding boxes, object classification, alerts) from the ingestion service.
Optimize Camera Handling
Replace thread-based camera handling with a GPU-optimized pipeline.
Pipeline flow: Camera → GPU Ingestion Service → Redis (metadata) → CPU Consumer (logic/alerts).
Secure the APIs
Add authentication mechanisms (API keys, OAuth, IP whitelisting) to secure the APIs.
Scalability and Resource Allocation
Add more GPU instances if needed and optimize resource usage to support more cameras efficiently.
System Reliability and Failover
Implement failover mechanisms (multiple GPU instances) to ensure the system stays stable and avoids downtime.
Post-Feature Implementation Tasks
After core features (crossings, age/gender detection), optimize code and refine GPU/CPU handling to improve efficiency and scalability.
Can you help us with this?