What is Orchid AI?

Beginning in version 25.9, Orchid VMS introduces Orchid AI Search. Users can search for specific objects that may appear using the familiar Orchid Smart Search and Thumbnail Search interfaces.

When AI Search is active, the search results will appear along the top of the Timeline Bar (just like Smart Search events) so they are easy to identify. You may then use the Jump To buttons to skip quickly from event to event. 

Orchid AI must enabled on a per-camera basis, and requires an AI License.  Orchid users with a Trial or Fusion license but no AI Licenses may configure AI Search on one camera and will have access to 24 hours of AI Search results.

Orchid Hybrid SaaS customers receive AI Licenses for all of their cameras at no additional charge.

Enabling Orchid AI

Orchid AI must be enabled in the settings of each camera for which you wish to access AI Search results.  Note that AI Search has server processing requirements that may exceed the capabilities of your current Orchid Recorder system.  See Performance Considerations below.

Once enabled, AI Search will be immediately available.  Note, however, that AI Search results are generated during the recording process and so they will not be available for periods of recordings during which AI Search was not enabled.

Performance Considerations

By default, Orchid AI will use your Orchid Recorder server's CPU to decode and perform AI analysis.  This is a computationally intensive process and a typical Orchid Recorder server may only have capacity to use AI Search on one or two cameras.

Orchid AI analysis runs at a lower priority on your Orchid Recorder system than the rest of the VMS, which means that even if you enable AI Search on more cameras than your system can support, core VMS functionality will not be significantly impacted, although the analysis framerate for AI Search will drop precipitously.  System administrators are encouraged to monitor the CPU and memory usage on their Orchid Recorder servers after enabling AI Search.
 

Optimizing AI Search Performance

Administrators can take several steps to maximize the number of AI Search streams usable on their Orchid Recorder system:

• Configure your primary stream to use a lower resolution: AI Search processing is performed on your cameras' primary streams and produces diminishing returns for resolutions greater than 640 x 480.  If possible, use a lower resolution primary stream and configure a secondary stream for your desired maximum resolution.
• Reduce your primary stream framerate: By default, AI Search is performed on every video frame.  Many users report excellent AI Search results at 5 frames per second, if not lower.
• Perform AI Search only on key frames: Orchid AI can be configured to only analyze video key frames (also called I-frames).  These video frames are uniquely inexpensive to decode but only occur every one to two seconds.  Enabling this capability will increase the number of AI Search cameras your system can support at the expense of lower detection framerates.  In the Orchid properies file, set 
  analytics.pipeline.key_frames_only: true and restart the orchid_analytics service.

Configuring GPU Acceleration

Orchid Recorder systems with a compatible NVIDIA GPU can be configured to use GPU-accelerated AI processing to significantly increase the number of cameras on which AI Search may be enabled.

Prerequisites

In order to use GPU accelerated AI Search processing, Orchid Recorder systems must:

• Run a supported Linux operating system.  This includes Ubuntu >= 18.04 LTS, RHEL >= 8, and recent versions of Fedora or SuSE.
• Have an NVIDIA GPU with Compute Capability >= 7.5.
• Have an installed NVIDIA driver with version >= 530.
• Install the Orchid VMS CUDA Support Package matching its Orchid Recorder version number.  Note that if automatic updates are configured within Orchid Recorder, the Orchid VMS CUDA Support Package will be automatically updated, as well (after initial installation).

Once all prerequisites are satisfied, GPU acceleration may be enabled by setting the following property in the Orchid Recorder properties file: analytics.object.execution_provider: cuda and then restarting the orchid_analytics service.

Administrators may monitor the GPU load and memory performance on their Orchid Recorder NVR by using a utility such as nvidia-smi.