A user computer’s network behavior changes every day; an IP camera’s stays almost the same for years. It talks to the same recording server, on the same ports, at about the same bitrate. This predictability makes the camera network one of the most suitable segments for anomaly detection. Writing the rule is easy; the hard part is the right baseline. Below we explain how to build a baseline of camera traffic with flow data, DHCP, DNS and switch logs, suspicious behavior indicators, example alarm rules and how to avoid alarm fatigue on the SIEM side.
Why camera traffic lends itself to monitoring
In a typical corporate installation, the systems a camera talks to are limited:
- Recording server (NVR/VMS): the RTSP session is usually initiated by the recording server (TCP 554). Media flows from camera to server either within the same TCP connection or over RTP/UDP.
- Infrastructure services: NTP, DHCP, sometimes DNS, syslog and SNMP.
- Management: HTTPS access from specific management stations.
- Event notification: in some installations, event delivery from the camera to the VMS or an SMTP server.
Any connection outside this list is either a misconfiguration or an incident to investigate. There’s no third possibility. On the user network a “new destination IP” appears every second; on the camera network it may not appear for months. That’s why even simple rules work with few false alarms. To define allowed flows in advance, also use the matrix in the IP camera traffic matrix and ACL design guide as a reference on the monitoring side.
How to build the baseline
- Define the scope: list camera VLANs, recording servers and management stations by name and IP. Define them as separate asset groups in the flow analysis tool.
- Collect data: cover at least a full week. Day/night bitrate differences (noise in night images can raise bitrate under VBR), weekends and maintenance days should be visible in the data.
- Classify flows: group them by source group, destination group, protocol and destination port. Mark expected flows as “allowed”.
- Sort out the unexpected: flows such as queries to public DNS servers, NTP to the internet and manufacturer cloud domains mostly come from default settings. It’s a frequent finding in the first week. Turn them off if unnecessary; if needed, document them explicitly.
- Build a per-camera profile: average and peak bitrate, number of sessions, number of connecting clients. For the logic of the bitrate calculation, see the camera network bandwidth calculation guide.
The baseline isn’t permanent. It should be reviewed when new cameras are added, after firmware updates or when the VMS server is moved. A baseline that isn’t reviewed has a predictable end: at the first big change, either the rules are muted or an alarm storm begins.
Data sources: NetFlow, sFlow, DHCP, DNS, switch logs
| Source | What it shows | Use on the camera network |
|---|---|---|
| NetFlow v9 / IPFIX | Per-flow source, destination, port, bytes | New destinations, new ports, volume changes |
| sFlow | Sampled packet headers and counters | High-volume trends; may miss short and small flows |
| Firewall logs | Permit and deny decisions | Attempts to leave the segment |
| DHCP logs | New MACs, new leases | An unknown device entering the camera VLAN |
| DNS logs | Queried domain names | Cloud service or command-and-control domains |
| Switch syslog | Link state, MAC moves, port security, 802.1X | Unplugged cables, device replacement, MAC cloning |
Where possible, take flow data from the L3 interface that is the camera VLAN’s gateway or from the segment firewall; Flexible NetFlow support on access switches depends on model and license. For the differences between flow protocols, see the NetFlow, sFlow and IPFIX comparison article on our network site. A basic Flexible NetFlow configuration for the camera VLAN interface on Cisco IOS XE (addresses are examples):
flow record CAMERA-REC
match ipv4 source address
match ipv4 destination address
match ipv4 protocol
match transport source-port
match transport destination-port
collect counter bytes long
collect counter packets long
!
flow exporter CAMERA-EXP
destination 10.10.5.20
source Vlan30
transport udp 2055
!
flow monitor CAMERA-MON
record CAMERA-REC
exporter CAMERA-EXP
!
interface Vlan30
ip flow monitor CAMERA-MON input
On the collector side, NetFlow v9 and IPFIX (RFC 7011) are template-based; the collector can’t decode records until it receives the exporter’s template. That’s usually why no data appears in the first minutes after a new exporter is added. It’s not a fault, it’s a wait. If data never arrives, the first place to look in the console is the UDP 2055 path between the exporter’s source interface and the collector; an ACL or firewall rule in between can silently drop this traffic.
Suspicious behavior indicators
| Indicator | Possible meaning | Severity |
|---|---|---|
| A connection attempt from a camera to the internet | A P2P/cloud setting left on, or a command-and-control connection | High |
| Connections from a camera to many IPs within the camera VLAN | Scanning, worm-like spread | Critical |
| Connections from a camera to TCP 23/2323 or 22 | Known spreading behavior of IoT botnets | Critical |
| RTSP or web access from a host other than the recording server | Unauthorized viewing or reconnaissance | High |
| A marked and lasting change in a camera’s bitrate | A settings change, an extra stream, data exfiltration | Intermediate |
| A new MAC address, or a camera’s MAC appearing on another port | Device replacement or MAC cloning | High |
| Unexpected DNS queries from a camera | A cloud service, an update check or a malicious domain | Intermediate |
| After-hours management interface logins and series of failed logins | Password guessing, unauthorized administration | High |
None of the indicators in the table is conclusive proof on its own; but on a camera network each deserves investigation. Especially if connections from a camera to many IPs within the camera VLAN are seen together with traffic to Telnet ports, don’t wait.
Example alarm rules
The rules below are written as product-independent logic; translate them into the query language of your own SIEM or flow analysis tool. CAMERA_NET, RECORDING_SERVERS and MANAGEMENT represent defined asset groups.
RULE-1 Internet egress
source IN CAMERA_NET AND destination NOT IN RFC1918
AND destination NOT IN (224.0.0.0/4, 255.255.255.255, 169.254.0.0/16)
-> alarm on every event (including firewall deny logs)
RULE-2 Lateral scanning
source IN CAMERA_NET AND destination IN CAMERA_NET
AND distinct destination IPs in 5 minutes > 10
-> critical alarm
RULE-3 Unauthorized viewer
destination IN CAMERA_NET AND destination_port IN (554, 80, 443)
AND source NOT IN (RECORDING_SERVERS, MANAGEMENT)
-> high alarm
RULE-4 Unknown device
DHCP lease VLAN = camera AND MAC NOT IN camera_inventory
-> high alarm
RULE-5 Bitrate deviation
per-camera 1-hour average bitrate
> 2x baseline OR < half of baseline (if it lasts 30 min)
-> medium alarm, correlate with camera health alarms
The multicast, broadcast and link-local exception in RULE-1 keeps local traffic such as WS-Discovery, SSDP and DHCP from producing false alarms. The thresholds are examples; adjust them to the real distribution in your baseline. RULE-5’s lower threshold is also a fault symptom: a covered lens, a frozen camera, a stuck encoder. For such availability and video loss alarms, use the camera health monitoring: SNMP, ONVIF events and video loss alarms guide. Correlating the two monitoring lines in the same incident record prevents unnecessary duplicate alarms.
SIEM integration and alarm fatigue
- Asset context: import the camera inventory (IP, MAC, model, location, owner) into the SIEM. An alarm without a location makes the field team first ask “which camera?”; response time is lost there.
- Maintenance windows: firmware updates and bulk configuration work are defined as suppression windows through change records. But suppression should never cover the most serious rules, such as “internet egress”.
- Aggregation and grouping: repeated events from the same camera are gathered under a single alarm. Behavior seen on many cameras at once, however, is escalated separately; it’s either a bulk configuration error or a spreading attack.
- Ownership: who does the alarm go to? If the camera system belongs to a contractor and the network to IT, routing rules must be written in advance. An alarm with no owner never gets closed.
- Retention: the retention period for flow and log data is set according to incident investigation needs; since records may contain personal data, it must comply with the organization’s policy.
First steps when an incident is suspected
- Verify: put the flow that raised the alarm, the camera’s inventory record and recent change records side by side. If it’s explained by planned work, close it and review the rule.
- Isolate, don’t shut down: instead of unplugging the camera, move the switch port to a quarantine VLAN or cut its access with an ACL. As long as the device stays on, the connections it makes can be observed and, if the manufacturer supports it, a system log or diagnostic package can be taken. Rebooting wipes these volatile traces.
- Capture the traffic: redirect the camera’s traffic to an analysis station with SPAN/port mirroring on the switch and take a short packet capture.
- Determine the scope: look for the same indicator on other cameras of the same model or firmware; check recording server and management station logs. A single camera rarely stays single.
- Clean up and harden: reinstall the firmware using the manufacturer’s recommended method, renew all credentials and certificates, disable services that were opened; only put the camera back into production under monitoring.
To add flow monitoring, alarm rules and SIEM integration on the camera network to your existing monitoring infrastructure, see our camera network monitoring and fault alarms service; for segmentation and service hardening, see the camera cybersecurity hardening page.
Frequently asked questions
Is a separate NDR product required for anomaly monitoring on the camera network?
It isn’t required. Because camera traffic is very predictable, NetFlow/sFlow data from switches or firewalls, DHCP and DNS logs and simple rules written in the SIEM catch most of the important indicators. NDR products offer scale and automatic profiling advantages.
How long should it take to build the baseline?
It should cover at least one full weekly business cycle; data should be collected until day and night differences, weekends and planned maintenance periods are visible. After that, the baseline is reviewed after every change and firmware update.
Is a camera trying to reach the internet always an attack?
No. A P2P cloud setting left on, a firmware update check or a public NTP server setting also produces this behavior. However, if the organization doesn’t need this traffic, the setting should be turned off and every remaining attempt investigated.
Is it right to shut down a suspicious camera immediately?
It’s usually better to isolate it first. Shutting the camera down wipes volatile traces on the device; moving the port to a quarantine VLAN or cutting access with an ACL stops the spread while leaving traffic records and the possibility of investigation.