Direct answer: “AI by NVR” means the recorder analyzes compatible incoming video; “AI by camera” means the camera creates the event or metadata and the recorder receives it. Camera-side processing usually scales with the number of intelligent cameras, while recorder-side processing can add analytics to selected compatible channels. Exact channel limits and feature combinations vary by model.
Hikvision AcuSense NVR Guide: AI by NVR vs AI by Camera
AcuSense is commonly associated with filtering events around people and vehicles. The label alone, however, does not tell you how many channels are analyzed, where processing happens, or whether two intelligent functions can run together. Those details determine whether the design meets the site’s actual response workflow.
The two AcuSense processing paths
Path A: AI by NVR
The recorder applies supported analytics to selected compatible streams. This can be useful when retaining conventional cameras, but the recorder has a finite AI workload.
Verify: supported channels, resolution/codec restrictions, event type, simultaneous functions and whether enabling one feature reduces another capacity.
Path B: AI by camera
The camera performs supported classification or detection and sends events or metadata to the recorder. Capacity is distributed across the camera estate, but camera model and firmware compatibility become critical.
Verify: camera event support, metadata handoff, NVR event display/search, firmware pairing and licensing in any management layer.
Some systems use both paths. Treat them as a channel-by-channel design decision rather than a global switch.
Compatibility matrix for the quotation
| Design question | Why it matters | Evidence to request |
|---|---|---|
| Which channels use AI by NVR? | Recorder analytics capacity is model-specific | Datasheet line plus channel allocation schedule |
| Which channels use AI by camera? | Event quality and metadata depend on the camera | Exact camera models, firmware and supported event list |
| Is AcuSearch required? | Search support is not identical across all AcuSense NVRs | Recorder/camera pairing and a recorded-video test |
| Are facial functions required? | Capacity, privacy obligations and feature concurrency differ | Written use case, legal review and model-specific capacity table |
| Are third-party cameras included? | Basic ONVIF video does not guarantee proprietary analytics | Registered conformance, event mapping and proof-of-concept result |
Use a workload schedule, not an “AI NVR” label
Make one row for every camera. Record the scene, resolution, frame rate, codec, event type, processing location, retention and required search method. A perimeter camera may need line crossing with person/vehicle filtering, while an entrance camera may only need dependable motion review. Allocating the same analytic to every channel wastes capacity and complicates testing.
- Define the response. Who receives the event, and what action follows?
- Choose the detection point. Place the analytic where the scene and hardware support it.
- Reserve headroom. Do not size the NVR at its theoretical limit.
- Specify evidence retrieval. An accurate event is less useful if operators cannot find the clip quickly.
- Test day and night. Classification performance depends on target size, angle, lighting and occlusion.
Current model examples: compare families carefully
The HIKD catalog includes different recorder generations and capacities. Examples such as the DS-9664NXI-I8/S, DS-7708NXI-K4 and DS-7732NXI-K4/16P should not be treated as interchangeable solely because their names include N, NXI or an AcuSense reference.
On newer product pages, Hikvision separates AI capabilities processed by the recorder from those received from compatible cameras. For example, the official DS-7608NXI-I2/S product page presents these categories independently and identifies hardware revisions. Always match the regional datasheet to the quoted suffix.
Camera selection matters just as much. A model such as the DS-2CD3686G2T-IZS may form part of an AcuSense design, but the installer still needs to validate lens choice, mounting geometry and the event passed to the recorder. Browse the wider NVR range only after defining those channel requirements.
AcuSense acceptance test
Use a recorded test sheet with pass/fail criteria. Run it after final mounting and again after major firmware or scene changes.
- Walk a person across the detection zone at near, middle and far distances.
- Drive the expected vehicle types through the zone at normal speed.
- Introduce common nuisance motion: foliage, rain, shadows, headlights and small animals where safe.
- Confirm the NVR displays the correct event, camera and timestamp.
- Search for the event using the operator’s real account and document retrieval time.
- Test notification, audio warning or other linkage without causing unsafe escalation.
- Repeat in representative low-light conditions and after switching illumination modes.
Questions buyers often ask
Can an AcuSense NVR make any camera intelligent?
No. Recorder-side analytics apply only to supported streams and a model-specific number of channels. Proprietary camera-side events require compatible cameras.
Does ONVIF guarantee AcuSense event support?
No. ONVIF profiles define specific interoperable functions, but they do not automatically reproduce every manufacturer-specific analytic. Verify the exact registered profile and event behavior.
Should analytics run in the camera or NVR?
Use camera-side processing when the camera and project support the required event at scale; use NVR-side capacity selectively for compatible conventional streams or centralized functions. The best design may combine both.