Data Storage Considerations

Before diving into the specific approaches, it's important to understand the data storage requirements for security cameras:

• A typical IP camera with 2000 kbps bitrate generates approximately 20 GB of data per day
• With optimizations like motion-based recording, this can be reduced to 10-15 GB per day
• Most businesses require 30-day storage, though some industries need 60-90 days
• Storage location affects cost, security, and accessibility

Cloud-Only Solutions

Pros:

• No on-premises hardware to maintain
• Easily accessible from anywhere
• Automatic updates and feature additions
• Simplified deployment
• Potentially unlimited storage capacity

Cons:

• Significant storage costs: Cloud storage can cost approximately $0.023 per GB per month, which adds up to $10+ per month per camera for 30 days of storage
• High bandwidth requirements (15+ GB per day per camera)
• Depends on reliable internet connection
• Ongoing subscription costs
• Data security and privacy concerns
• Latency issues for real-time AI compute
• Limited by upload bandwidth constraints

Hybrid-Cloud Solution

Hybrid-cloud approaches that provide the best of both worlds:

• Low-cost, low-bandwidth video storage on-premises
• Optional cloud backup for important cameras
• Unlimited cloud archives for critical video clips
• Flexibility to choose storage duration (30/60/90+ days)
• AI compute can be scaled with local hardware

Closed Camera (Hybrid Cloud)

example:

Pros:

• camera is fully self-sufficient, you don’t need any middle man such as NVR
• for smaller locations, you have to justify the cost of NVR.
• we control the hardware
• fast firmware upgrades

Cons:

• if we choose this approach, we are creating a lock in mechanism that makes it hard for using existing cameras
• you have to work with SD cards which, could be removed from the cameras at any given point.
• we can end up in a situation where you need to replace camera to get additional features
• camera needs to be more expensive when it needs AI chips and not only ONVIF compliance.
• Needs firmware engineering experience or SDKs.

Examples: Verkada, Rhombus Systems

Port Forwarding (Cloud Only)

This approach involves configuring a router to forward specific ports to the camera, enabling direct access to the camera's RTSP stream from the internet

Pros:

• No additional hardware required beyond existing cameras and router
• Works with a wide range of camera models that support RTSP (ONVIF)

Cons:

• May not work with all types of internet connections (carrier-grade NAT issues)
• Requires ongoing management of dynamic DNS
• Some ISPs or corporate networks may block necessary ports
• Scalability issues as each camera needs its own unique external port
• Introduces latency for AI-heavy features

Example: CamCloud, Reolink

AI NVR (Hybrid Cloud)

• Storage control: you have the full control over storage →can sync/upload to cloud with full timelines
• Streaming quality: you can have higher quality streaming
• AI compute: generally we should have sufficient hardware for 8 cameras with some reserve in compute for expensive application. If you need more cameras or compute, you can buy additional hub or rack.
• Existing IP Cameras: If you already have ONVIF-compliant IP cameras, then using an NVR that can support any IP camera will be more cost-efficient.
• Pricing: Closed camera systems are typically more expensive than ONVIF-compliant IP cameras
• AI NVRs offer better upgrade paths without replacing cameras

Cons

• You have to justify higher up front cost

Examples: Coram Point, Coram AI

Hardware Options:

• Mini: Supports up to 4 camera streams
• Medium: Supports up to 8 camera streams
• Large (L1): Supports up to 16 camera streams
• Large (L2): Supports up to 24 camera streams
• Large (L3): Supports up to 32 camera streams

Core Features of Modern AI Video Surveillance

Advanced systems should offer:

• Multi-camera playback from both local and cloud storage
• Easy switching between recorded footage and live feeds
• Metadata overlay on video timeline
• Search capabilities by face, time, and attributes
• Future capabilities like text-based search and person tracking
• Data protection options
• Efficient bandwidth usage

Conclusion

When selecting a video surveillance approach, consider:

1. Existing infrastructure: If you already have ONVIF-compliant IP cameras, an AI NVR approach is more cost-efficient
2. Scale: For small deployments, closed camera systems may be simpler, while larger installations benefit from AI NVRs
3. Computing needs: AI capabilities require substantial computing power, which AI NVRs provide more abundantly
4. Future-proofing: AI NVRs offer better upgrade paths without replacing cameras
5. Cost structure: Balance upfront costs against ongoing subscription fees
6. Data security: Consider where your data is stored and who has access to it

‍