Platform 4.0

Reference architecture

Separate sources, execution, data management and external consumers.

Browsing: Platform 4.0
Also available in: Platform 4.0BOX 5.0

Sources and edge processing

Cameras, VMS and other sources connect to a managed execution layer. Decide where decoding and inference run using latency, connectivity and data-handling requirements. The source must supply an image suitable for the intended analytical task.

Inference and scenarios

Models produce perception results. Scenario graphs combine those results with tracking, zones, external signals and operating rules. Model and scenario versions are distributed in a controlled way; consumers should not depend on undocumented model internals.

Data and control

Operational metadata, analytical records and media have different storage profiles. The documented architecture includes PostgreSQL for metadata, object storage for packages and media, and messaging between services. The exact service set and redundancy arrangement are defined in the delivery.

Integration boundary

External systems receive agreed event and reference-data contracts. Define source identifiers, timestamps, evidence references, verification state and retry behaviour independently of the implementation of a particular detector.

Operational checks

  • Trace one source through frame acquisition, inference, scenario processing, event storage and operator review.
  • Record the versions and service owners at each boundary.
  • Monitor stream health, processing delay, storage and failure/retry behaviour.
  • Validate the selected topology on target hardware and representative data.

Test-run storage and execution lifecycle

An asset test and continuous camera processing have different lifecycles. The test path selects entity and asset versions, obtains files/zones and creates launch_uuid; a Sandbox execution session runs the model or graph and returns results. Registry metadata is separate from the processes that execute the model.

Store / transportRole in model and run processing
PostgreSQLRecords, versions, run states and result references
MinIO / S3Model packages, assets, original/visualized frames and logs
RedisLive execution-session state in the applicable component
KafkaVersion notifications, continuation tasks, progress and state changes

Preserve the relationship between metadata and stored objects during backup and diagnosis. Restoring a database without the corresponding model files and results does not restore the whole working system.

Use the documentation and configuration supplied for your installed release. Environment addresses and access credentials are provided privately with the delivery.