Definition
VMS (Video Management System) centrally manages IP cameras and NVRs, providing video storage, playback, and intelligent analysis. Video edge analysis devices support RTSP, ONVIF, GB28181 camera access and report alarms to the VMS or third-party security platforms via MQTT/HTTP.
Core Functions
VMS provides device discovery and access (ONVIF, RTSP, GB28181); real-time preview and multi-screen display; video storage, retrieval, and playback; PTZ control and preset positions; alarm management and event linkage; user permissions and operation logs. It supports integration with access control, fire protection, BMS, and other systems. Video edge analysis devices (AI boxes) can pull camera video, run local object detection and behavior analysis, and report alarm events to the VMS, achieving integrated intelligent analysis and video management.
Relationship with GB28181
GB28181 is the national standard video networking protocol. Platforms such as public security and traffic systems require devices to access via GB28181. VMS can act as a GB28181 platform to receive cascaded video, or convert ONVIF devices via a gateway to access the upper-level platform using GB28181. Commercial VMS often need to support public security platforms simultaneously.
Recording Index and Alarm Choreography
Devices, streams, and recordings belong in one directory
A video management system maintains a directory, not only a grid of pictures. Each camera in the directory has to be bound to a channel number, a way of pulling the stream, a storage location, and who may watch it. Live view, recording playback, and alarm pop-ups all take the identity of the same path from this directory. If the directory and the storage each use their own names, an alarm opens and the recording is missing, or it opens the neighboring channel. An index has to record at least the start time, the end time, and where that file sits on disk. Search can then jump by time, instead of making someone open files one by one.
Retention in days is set together by bitrate, path count, the hours actually recorded each day, and usable disk capacity. Motion-only recording writes only when there is activity, so retention grows longer, but a sudden increase in activity shortens it. Plan disk for the peak, not for a quiet night. The overwrite policy should say whether the oldest recording is deleted or recording stops. Stopping means there is no picture on the day of an incident. Deleting the oldest means a clip still under investigation has to be locked.
Estimate retention from bitrate
The conditions for use are a camera clock, a server clock, and an index clock that agree, permissions cut by role, and a storage path that raises a clear alarm when a disk fails. It suits a site where several people preview at once, evidence is retrieved by time, and door access or alarm inputs are bound to a picture. A stretch of local file with nobody and no event to align does not need a full directory. If analytics finish on an edge box, the system should receive an event that carries a channel number and a time, then align it with the local recording. Do not store a second full-quality copy of the decoded stream that analytics used.
A linkage has to state the trigger and the action. Which path pops up when a discrete input arrives, whether a preset is called, and whether the recording is locked, are rules. They are not something the camera brings with it. If a channel a rule cites is deleted or renamed, the rule should fail and alarm. It should not point silently at a different camera that was connected later.
Clocks, permissions, and a duplicated cascade
Clocks that disagree are the first reason playback does not meet the event. A camera a few minutes slow, with the event recorded in server time, leaves the search window empty, and people conclude it was never recorded. Force time sync at connection, and refuse to admit a device whose time has jumped too far. Second is permission. If preview, playback, and export are bound to the same administrator account, any terminal can copy the recording away. An export should keep the operator, the channel, and the time range.
Third is a duplicated cascade. The same path is recorded by the local system and pulled again as a main stream by the upper platform. The camera's sessions fill up and both sides stutter. Agree who stores and who only watches the stream. If there is no early alarm when the disk is nearly full, overwrite is discovered only after the incident. After a disk is replaced, if the index is not rebuilt, the directory still points at the old drive letter and every playback fails. Finally, do not substitute a time burned onto the picture for the index time. The overlay can be covered, or burned in the wrong time zone. Evidence still follows the index.
Rehearse recording lock and the overwrite policy together. A lock clicked in the interface, while the disk task still deletes by time, means the clip is already gone when the investigation starts. Fill one path to the watermark, confirm that locked clips are kept and only unlocked clips are overwritten, and confirm the alarm reaches the duty desk before the watermark arrives. Without that rehearsal, a capacity plan is only a static table.