Glossary / Protocols

What is RTSP?

RTSP is a network control protocol that lets software start, pause and stop a video stream from a camera or recorder. It carries the commands, not the video itself, which travels over RTP. Almost every IP camera and NVR exposes RTSP, which makes it the standard route for adding analytics to hardware already installed.

Reviewed by Dr. Vivek Gujar, Chief Strategy Officer  ·  Updated 31 August 2026  ·  6 min read

RTSP is often described as a way to stream video. That is close enough for everyday use but slightly wrong in a way that matters when things break. RTSP is a control protocol. It negotiates the session and carries the commands: describe, setup, play, pause, teardown. The video and audio travel separately over RTP. When people say a camera has an RTSP stream, they mean it accepts RTSP control and delivers RTP media.

Its importance in Indian deployments is practical rather than technical. Most sites that want analytics already have cameras. Those cameras are frequently five or eight years old, from a mix of vendors, feeding a recorder that works. RTSP is the interface that lets new software read that video without touching any of it.

What an RTSP URL contains

Four parts matter. The credentials, which is where the first security problem lives. The address and port, conventionally 554. The path, which is vendor specific and undocumented as often as not. And the stream selector, buried in that path, which chooses main or sub stream.

There is no standard path format. Every manufacturer invents its own, which is why integrators keep lists of URL patterns per brand and why ONVIF discovery is worth having: it asks the camera for its stream URIs instead of guessing them.

Main stream and sub stream

Almost every IP camera publishes at least two encodings of the same video simultaneously. This is the single most useful fact about RTSP for anyone doing analytics.

Typical dual stream configuration and what each is for

Stream	Typical settings	Bandwidth	Use
Main	1920x1080 at 25 fps	2 to 4 Mbps	Recording and evidence. Retained for review and export.
Sub	640x360 at 10 fps	0.3 to 0.5 Mbps	Live preview, and detection for most analytics tasks.

Running detection on the sub stream costs roughly one eighth the bandwidth and a similar fraction of the decode load, which is why a modest appliance can handle many cameras. Person and vehicle detection work perfectly well at 640x360. Tasks that need fine detail do not: plate reading and face work need the main stream or a dedicated camera, because pixel density at the target is the limiting factor.

The usual pattern is to detect on the sub stream and pull the main stream only for the clip around a confirmed event. Running every analytic on every main stream is the most common reason a retrofit appliance is undersized.

Pulling from the NVR or from the camera

Both work, and the choice has consequences.

• From the camera. Lower latency, one less dependency, and full control of stream settings. But every consumer counts against the camera's concurrent connection limit, which on older models can be as low as four including the NVR itself.
• From the NVR. One device to configure and credential, and the NVR is already reachable. But it adds latency, and NVR RTSP re-streaming is frequently more limited than the camera's own, sometimes capped at fewer simultaneous outputs than the channel count.

Check the concurrent stream limit before designing. It is the constraint that most often surprises people mid-deployment.

TCP or UDP

RTP can travel over either. UDP has lower latency and drops frames under congestion. TCP retransmits, so nothing is lost, but a congested link produces growing delay rather than dropped frames. For analytics, TCP is usually the right default, because a missing frame is a missed detection while a slightly delayed frame is not. On a clean local network the difference is negligible either way.

Security

Use per-consumer accounts rather than sharing the admin credential with every system that reads a stream. When a system is decommissioned you can then revoke one account instead of rotating the camera password and rediscovering everything that used it.

Frequently asked questions

Is RTSP the same as RTP?

No. RTSP is the control protocol that starts, pauses and stops the session. RTP is the transport that actually carries the video and audio packets. They are used together, and most people say RTSP when they mean both. The distinction matters when troubleshooting, because a session can negotiate successfully over RTSP and still deliver no media if RTP is blocked.

Do all IP cameras support RTSP?

Almost all do, but some consumer cloud cameras deliberately disable it to keep users inside the manufacturer's app. Before buying for a site where analytics may be added later, confirm that RTSP is exposed and not locked behind a paid tier. Cameras without accessible RTSP are difficult to integrate with anything the manufacturer did not build.

Should analytics use the main stream or the sub stream?

Sub stream for most detection tasks, because person and vehicle detection work well at 640x360 and the bandwidth and decode cost is roughly an eighth of main. Use the main stream where pixel density at the target is the limiting factor, such as plate reading, or pull it only for the clip around a confirmed event.

How many RTSP streams can one camera serve at once?

It varies by model and firmware, commonly between four and twenty concurrent connections, and the NVR already occupies one. Older cameras can be as low as four. Check the limit before designing a system where multiple consumers read the same camera, because exceeding it causes intermittent failures that are hard to diagnose.

Can I add AI to old cameras using RTSP?

Yes, and this is the normal retrofit path. An on-site appliance pulls RTSP from the existing cameras or NVR and runs the models locally, with no change to the recording setup. The practical limits are whether the cameras deliver enough resolution and frame rate for the task, and whether the concurrent stream limit allows another consumer.

Why does my RTSP stream keep dropping?

The three usual causes are exceeding the camera's concurrent connection limit, network congestion on a UDP transport causing packet loss, and credential or session timeouts on cameras that expire idle sessions. Switching the transport to TCP resolves a good share of intermittent drops on busy networks.