A thermal camera module becomes truly useful when it streams clean, low-latency video into phones, VMS software, or edge boxes. Whether you start from a thermal imaging camera core or a compact thermal image sensor module, this guide shows how to wire protocols, choose ONVIF/RTSP profiles, set encoder/bitrate, and verify glass-to-glass latency—using the same discipline we apply across Thermal camera module products and integration kits on Module Integration for OEMs.
Executive Summary
For dependable streaming, make four decisions explicit: (1) transport (RTP/UDP for lowest latency; RTP/TCP for lossy links; HTTP-tunneled RTSP only as a last resort); (2) ONVIF profile (S for basic H.264, T for H.265/eventing/metadata); (3) encoder ladder (resolution, frame rate, bitrate, GOP); and (4) time & discovery (WS-Discovery + NTP). Publish a one-page acceptance card and store golden PCAP/RTSP logs for each release under Downloads.
Use Cases & Buyer Scenarios
Scenario 1 — Handheld to mobile app (Wi-Fi STA/AP)
Users watch live thermal on a phone while recording snapshots. Prioritise <250 ms glass-to-glass latency, steady 8–12 fps for 640×512 (mobile decoders), and clean reconnect after RF drops. Pair with product families listed under Products.
Scenario 2 — Security VMS via ONVIF
Integrators want plug-and-play discovery, time sync, and PTZ/metadata. Use ONVIF Profile S or T with digest auth, WS-Discovery, and RTSP over UDP by default. Provide a conformance sheet and sample XMLs in Support.
Scenario 3 — Edge AI box / vehicle network
Deterministic feeds to an edge computer on 100BASE-TX/PoE. Fix frame cadence, lock GOP, and stamp monotonic PTS. If your rig also carries a rangefinder, follow our fusion notes at Thermal + LRF Fusion & Ballistics.
Spec & Selection Guide (the heart)
Key parameters (definitions + why they matter)
| Parameter | Definition | Typical target | Why it matters |
|---|---|---|---|
| Transport | RTP over UDP/TCP; RTSP control channel | UDP default; TCP fallback | UDP cuts jitter; TCP survives Wi-Fi/4G loss |
| ONVIF profile | S (H.264), T (H.265/analytics), G (recording) | S or T for live | Interoperability with VMS and NVRs |
| Encoder | H.264/AVC or H.265/HEVC; MJPEG for debug | H.264 baseline/main for phones | Trade quality vs compute and power |
| Bitrate (CBR/VBR) | Target kbps + buffer; GOP = I + (P/B) | 640×512@30: 1.5–3 Mbps (H.264) | Too low → banding/blocking; too high → drops |
| GOP / keyframe interval | I-frame period in frames or seconds | 1 s (e.g., 30 at 30 fps) | Recovery after loss; seek speed |
| Color/AGC | 14-bit → 8-bit mapping; palette | Local AGC, iron/white-hot | Legibility and encoder efficiency |
| Latency budget | Sensor→ISP→encode→network→decode→display | <250 ms handheld; <400 ms VMS | Human perception & control loops |
Quick selection rules
- If the viewer is a phone/tablet, then prefer H.264, 640×512 @ 8–15 fps, 1–2 Mbps, GOP 1 s, RTP/UDP with TCP fallback.
- If the viewer is a VMS/NVR, then expose ONVIF Profile S/T, CBR 2–4 Mbps, multicast on LAN, WS-Discovery + NTP.
- If the link is lossy or over WAN, then switch to RTP/TCP, shorten GOP (0.5 s), and enable error concealment.
Integration & Engineering Notes
Electrical & Interfaces
Ethernet/PoE. 10/100BASE-TX with 802.3af PoE keeps wiring simple; spec budget headroom (module + encoder + heater). Keep PHY layout away from sensor analog, and shield the RJ45 shell to the chassis. Wi-Fi. For handhelds, STA mode to a known SSID or AP mode for direct connect; hide the RTSP port behind a QR-code URI in your app. USB. If your core exposes UVC/YUY2 or H.264, treat it as a debug path; RTSP over Ethernet scales better.
Optics & Mechanics (thermal-specific)
NUC (shutter) pauses cause visible frame stalls—announce them in the RTSP metadata or desensitise the encoder during a NUC. Keep the window hydrophobic and AR-coated for legibility. If your device also houses a rangefinder, align the boresight and overlay discipline described in Thermal + LRF Fusion & Ballistics.
Firmware/ISP/Tuning
- AGC. Use local AGC with clipping guards; stable histograms compress better and reduce pumping.
- Palette. White-hot or iron compresses efficiently; avoid rapid palette changes during streaming.
- Encoder ladder. Offer at least two: 640×512@8–15 fps (1–2 Mbps) and 320×256@15–30 fps (0.3–0.8 Mbps). H.265 halves bitrate but increases decode cost.
- GOP & VBV. Set keyframe every 1 s; VBV buffer ≈ 1–2× target bitrate for smoothness; prefer CBR for VMS.
- Timestamps. Stamp PTS/DTS monotonically and sync clock via NTP; expose timezone and DST in ONVIF Device service.
- Security. RTSP digest auth; disable basic auth; rotate credentials; allow TLS interop if your stack supports RTSPS.
Testing & Validation (bench → field)
Acceptance targets (illustrative). Glass-to-glass <250 ms (handheld) or <400 ms (VMS); packet loss <0.5% over 5 min; re-connect <3 s after AP loss; keyframe interval within 10%; CPU headroom >20% at max profile.
Latency. Film a flashing LED with the module and the phone screen in the same video; measure frame offset. Robustness. Walk-test Wi-Fi; pull LAN and re-connect; check ONVIF events for “network lost/restored.” Bitrate/quality. Sweep 0.5→4 Mbps and log blocking/blur; verify VBV stability. Time/Discovery. Confirm WS-Discovery shows the device; NTP sync within 1 s over 24 h. Interoperability. Test with VLC, ffplay, Milestone/Exacq (Profile S), and an ONVIF Device/Media test tool; archive PCAP/RTSP DESCRIBE/SETUP/PLAY sequences in Downloads.
Compliance, Export & Certifications
Streaming products ship with EMC and safety marks (CE/FCC/UKCA), RoHS, and—where applicable—ONVIF conformance declarations. For export, many uncooled cores fall under dual-use rules (e.g., EAR 6A003/6A993 categories); check your jurisdiction before shipment. Maintain privacy features (passwords, disable default creds, opt-in metadata). Public certs and warranty terms belong on Certificates and Warranty.
Business Model, MOQ & Lead Time (OEM/ODM)
We ship thermal streaming kits with pre-set RTSP/ONVIF profiles, QR-code URIs, and sample apps. Typical MOQs: 100–300 pcs for catalog optics; 500–1,000 pcs for custom encoders/PoE boards. EVT 4–6 weeks (catalog); add 6–10 weeks for custom hardware. Explore bundles on Modules and families on Products.
| Option | Pros | Cons | Use when… |
|---|---|---|---|
| H.264 / RTP-UDP | Lowest latency; broad decode support | Packet loss shows as macro-blocks | Local LAN/Wi-Fi with reasonable SNR |
| H.264 / RTP-TCP | Survives loss; traverses firewalls | Higher jitter/latency | WAN links, 4G routers |
| H.265 / RTP-UDP | ~40–50% bitrate saving | Decode cost on low-end phones | Edge AI boxes; NVRs |
| MJPEG / HTTP | Debug-simple; intra-only | High bitrate; poor efficiency | Lab bring-up only |
Pitfalls, Benchmarks & QA
- Variable frame rate without proper timestamps. Causes audio/video drift in recorders; lock cadence.
- Huge GOPs (e.g., 4–5 s). Slow seek/recovery; cap at ~1 s for responsive UIs.
- “Unlimited” VBR. Peaks overwhelm Wi-Fi; prefer CBR or capped VBR.
- No NTP. Time jumps break recordings and signatures.
- Ignoring NUC impact. Shutter pauses look like freezes; flag them in metadata or during RTSP “info.”
- Default passwords. Blocked by many VMS/security buyers; force change at first boot.
- Wrong YUV/range. Crushes dark detail; test studio vs full range mappings.
FAQs
Q: What bitrate should I start with for 640×512?
H.264 at 1.5–2.5 Mbps and 8–12 fps gives a solid baseline for phones; increase to 3–4 Mbps for 30 fps or analytics.
Q: Which ONVIF profile should I claim?
Profile S covers H.264 live; Profile T adds H.265, events, and advanced metadata. Declare only what you test and publish a conformance sheet.
Q: Can I multicast?
Yes on LANs/NVRs; expose an IGMP toggle and document the group/TTL. Avoid multicast across Wi-Fi APs unless you control the network.
Q: How do I measure glass-to-glass quickly?
Record a flashing LED and the screen in the same video, then count frames; repeat at different bitrates and transports.
Q: Does HEVC always beat AVC?
Not on older phones or low-power decoders. HEVC saves bitrate but may raise latency/CPU. Offer both ladders.
Decision Flow — from bring-up to acceptance
Start
├─ Viewer? phone / VMS / edge box
├─ Transport: RTP/UDP default → TCP fallback → HTTP-tunneled as last resort
├─ ONVIF: Profile S (H.264) or T (H.265 + events)
├─ Encoder ladder: {res, fps, bitrate, GOP, CBR/VBR}; stamp PTS via NTP
├─ Network: WS-Discovery; digest auth; optional multicast (IGMP)
├─ Acceptance:
│ • latency < 250 ms (handheld) or < 400 ms (VMS)
│ • reconnect < 3 s; packet loss < 0.5%; keyframe every ~1 s
│ • NUC flagged; time drift < 1 s / 24 h
└─ Archive PCAP + RTSP trace + ONVIF XML under Downloads; release notes in Support
Call-to-Action (CTA)
Need a plug-and-play thermal stream? We ship cores and encoders with pre-tuned RTSP/ONVIF profiles, mobile demo apps, and acceptance scripts. Start a spec review via Contact, browse families under Thermal camera module, and see integration options on Module Integration for OEMs.
Sources
- RTSP 2.0 (RFC 7826). Control protocol for streaming sessions; DESCRIBE/SETUP/PLAY. (RFC Editor)
- RTP (RFC 3550). Real-time transport over UDP/TCP; timing & loss handling. (RFC Editor)
- ONVIF Profiles S/T/G. Device, Media, Events, Recording services; conformance tools. (ONVIF)
- H.264 / H.265 Standards. Compression families and profile levels. (ITU-T H.264; ITU-T H.265)
- IEEE 802.3af PoE. Power over Ethernet basics and budgets. (IEEE)
