In modern physical security, organizations often face a frustrating dilemma: they have invested heavily in high-performance camera ecosystems, but their systems remain largely reactive. Traditional motion detection floods security teams with false alarms, while true threat detection, license plate intelligence, and human intrusion recognition remain out of reach unless you replace your entire hardware infrastructure.

At Sentry AI, we took a radically different approach when building our middleware engine:

You do not need proprietary gateways, specialized on-premise hardware, or months of integration cycles. If your camera or Video Management System (VMS) can fire an HTTP webhook, we can integrate with it — seamlessly, securely, and blisteringly fast.

What Exactly Is a Webhook, and Why Does It Make Integrations So Fast?

If you are unfamiliar with the concept, think of a traditional API as polling: your server repeatedly asks the camera, "Did anything happen yet? How about now? Now?" This consumes excessive bandwidth, hits rate limits, and introduces unwanted lag.

A webhook, on the other hand, is an automated, event-driven HTTP callback — essentially a reverse API:

1. A camera detects motion, a line-crossing, or a vehicle entry at the edge.

2. The camera platform immediately fires a lightweight `HTTP POST` request payload to our designated endpoint.

3. Our server receives the event, validates it, and acknowledges receipt in under 20 milliseconds.

Because webhooks rely on standard HTTP/HTTPS protocols, we don't have to build custom device drivers from scratch. The moment your system can notify an endpoint, our pipeline takes over.

How We Handle Media Extraction: Two Real-World Architecture Examples

Capturing a webhook alert is only step one; feeding high-resolution, context-rich frames into our computer vision models without lagging or thrashing server resources is where the real engineering begins.

Here is how our middleware processes events from two premier systems: Avigilon Alta and Eyebuild.


1. Avigilon Alta: "Time-Traveling" Snapshot Intelligence

When an Avigilon Alta camera flags an incident, critical context often happens seconds before the trigger is recorded (e.g., an intruder sneaking into the frame before breaking a perimeter).

Our Alta integration handles this with an intelligent time-offset retrieval engine:

  • Pre-Roll Capture: Upon receiving Alta's motion webhook, our engine sets its fetch pointer to 5 seconds prior to the alert's timestamp.
  • 1 FPS Microsecond-Precise Fetching: A dedicated scheduler ticks every second, invoking Alta's snapshot API with microsecond timestamps (`timestampMicro`) to reconstruct the scene frame-by-frame.
  • Centralized Motion Tracking: A lightweight poller checks Alta's alert status every 3 seconds. When motion stops, the system captures an extra 5 seconds of post-event context before closing the session.
  • Pipelined Batching: Frames are bundled in batches of 4 and streamed straight to Redis cache, Google Cloud Storage (GCS), and Cloud Pub/Sub so downstream AI inference begins while the incident is still unfolding.


 2. Eyebuild: Zero-Disk, In-Memory Video Stream Extraction

Eyebuild AI alarm payloads deliver event metadata alongside pre-signed MP4 clip URLs or high-resolution snapshot links. To handle enterprise-scale event volume, writing video files to local server disks would create massive I/O bottlenecks.

Our Eyebuild worker pipeline solves this with a zero-disk, streaming architecture:

  • Instant Handshake: The webhook is acknowledged in `<20 ms` and enqueued into a distributed task queue (`Asynq` backed by Redis), eliminating timeout risks.
  • In-Memory Pipe Streaming: An isolated `ffmpeg` subprocess streams the pre-signed MP4 video directly to standard output (`stdout`) at 1 frame per second (`-vf fps=1 -f image2pipe -vcodec mjpeg`).
  • Byte-Level Delimiter Parsing: Our Go worker reads the stream in 32 KB chunks in memory, scanning for raw JPEG boundary markers (`0xFF 0xD8` Start of Image and `0xFF 0xD9` End of Image). No temporary files ever touch the server disk.
  • Immediate AI Dispatch with Fallback: As frames are extracted, they are batched and published immediately to downstream AI models. If a video clip is ever unavailable, the system automatically falls back to the snapshot URL.


 Why This Matters to You

  • If You Own or Manage Avigilon Alta or Eyebuild Cameras:

You don't need new hardware to get state-of-the-art AI. We can plug into your existing Alta or Eyebuild deployment in a matter of hours.

  1. Eliminate up to 90%+ of nuisance alarms caused by headlights, weather, animals, or swaying trees.
  2. Unlock proactive threat detection: Autonomous human/vehicle intrusion recognition, license plate intelligence, and instantaneous dispatch-ready alerts.
  3.  Zero hardware upgrades, zero downtime.
  •   If You Are a Camera Manufacturer or VMS Platform:

Does your camera firmware or cloud platform support webhooks?

  1. Partner with us to offer advanced AI detection to your customers instantly.
  2. Enhance your product value and generate recurring software value without spending years building in-house computer vision pipelines.
  3. Our modular architecture means an integration with your platform can be launched, tested, and validated in days.


Ready to Turn Passive Cameras into Proactive Sentinels?

Whether you're looking to upgrade your existing Avigilon Alta or Eyebuild deployment or you're a camera vendor seeking an AI technology partner, Sentry AI makes it easy to bring intelligent, real-time threat detection to existing camera systems.

If your system supports webhooks, let's explore what we can build together.