Direct answer
Use the lowest stream that preserves the target object at the required pixel size. Measure bitrate under real motion, then reserve capacity for reconnects, bursts and operator viewing.
Why this happens
AI accuracy may need detail from the main stream, while a substream can be sufficient for large objects or scene-state detection. Variable bitrate also rises during motion, exactly when alerts matter.
Size the ingest path for the inference job, then keep the original stream, derived metadata and rendered overlays distinct.
A controlled test
Test representative scenes and compare model quality, bitrate and decode cost for each camera profile.
Change one variable at a time. Keep the camera model, firmware, endpoint and account recorded; then test network reachability, protocol response, media transport and decoding as separate layers.
Use a dedicated view-only account and a trusted local diagnostic tool. Redact credentials, private addresses and identifying data before sharing output.
Diagnostic sequence
| Check | Action | Evidence of progress |
|---|---|---|
| Pixel need | Measure target size in the inference image. | The minimum useful resolution is known. |
| Bitrate | Observe idle and motion peaks. | Network sizing includes real bursts. |
| Profile | Compare main, substream and cropped inputs. | The lowest adequate source is selected. |
| Shared use | Account for recording and human viewers. | AI does not consume all camera sessions or bandwidth. |
Evidence to keep
Keep a per-camera table with scene, target pixel size, profile, p95 bitrate, inference quality and capacity headroom.
Boundary and safety note
Bandwidth savings are not useful if they invalidate the detection objective; verify outcomes before scaling the design.
For remote viewing, use a managed VPN instead of exposing RTSP or camera administration ports directly to the public internet.
SmartRTSP
SmartRTSP is a camera-focused RTSP and ONVIF viewer for Apple devices, Windows and Android. It fits direct viewing, discovery and multi-camera checks; keep a dedicated NVR or VMS when continuous recording, evidence export or centralized enterprise controls are required.
Frequently asked questions
Is a substream always enough for AI?
No. It depends on target size, scene and model; validate it against the real event.
Why measure motion bitrate?
Variable-rate cameras often produce their highest bandwidth when the scene changes.
Can a crop reduce bandwidth?
A gateway or smart camera may provide a region or derived stream, but its processing cost and loss of context must be tested.
Primary references
- NVIDIA — DeepStream architecture overview
- ONVIF — Profile T for advanced video streaming
- ONVIF — Profile M for metadata and analytics
Related SmartRTSP guide
Open related guideSize camera ingest and inference traffic from the pixels the model needs, not the camera’s maximum specification.