--- title: "Video Surveillance Systems" description: "IP camera and VMS design: storage sizing, network segmentation, hardened devices and retention periods that satisfy GDPR rather than just filling disks." url: "https://nixit.io/services/video-surveillance/" lang: "en" --- # Video Surveillance Systems We design and operate IP surveillance as infrastructure: sized storage, isolated networks, hardened cameras and defensible retention. Most failed surveillance systems fail at the moment they are needed. The cameras are recording, the screens look convincing, and then an incident happens and the footage is 720p at eight frames per second, the camera covering the relevant door was pointed at a car park, and recordings older than four days were silently overwritten because nobody sized the storage. The equipment worked exactly as installed. It was designed by someone counting cameras rather than asking what the footage was for. We approach surveillance the way we approach any other infrastructure: define the requirement, size the system against it, isolate it properly and operate it deliberately. ## Design before hardware The useful question is not how many cameras but what each one must let you do. Identifying a stranger's face, reading a number plate at speed, and observing that a delivery arrived are three different requirements with three different pixel densities, lens choices and lighting needs. A camera that satisfies the third will not satisfy the first, regardless of its megapixel count. We work through this per location: field of view, mounting height, expected lighting through the day, and whether the scene demands wide dynamic range because it faces a glass entrance. Number plate capture in particular needs a dedicated camera with a controlled angle and shutter, not a general overview camera pointed hopefully at a gate. Megapixels get oversold. Sensor size, lens quality and low-light performance usually matter more, and an 8 MP camera on a small sensor frequently produces worse usable night footage than a well-chosen 4 MP one, while consuming twice the storage. ## Recording and retention Storage is calculated, not estimated. Bitrate, camera count and retention determine capacity, and the result should be presented as a number you can check rather than a package tier. We size for the required retention with headroom, and use H.265 with smart codecs where the cameras support it, since the saving is substantial on mostly static scenes. Recording servers need the same treatment as any other server: monitored disks, RAID with a tested rebuild, and alerting when a camera stops recording. A camera that has been offline for three weeks is an extremely common finding, and it is only ever discovered when someone goes looking for footage that does not exist. Recording failure should page somebody the same day. Retention is a policy decision with legal weight, not a storage setting. We configure it explicitly, document why that period was chosen, and make deletion automatic. ## Treating cameras as the attack surface they are An IP camera is an embedded computer with a network stack, a web server and a long history of vulnerabilities. Large botnets have been assembled almost entirely from cameras and recorders exposed to the internet with unchanged credentials. Any camera deployment is therefore also a security project. We place cameras on a dedicated VLAN with no internet route, replace default credentials, disable unused services such as UPnP and P2P cloud features, and keep firmware on a maintenance schedule. Remote viewing goes through a VPN or a controlled gateway, never through a port forwarded to a recorder. Where procurement rules matter, we also check device provenance, since several widely sold brands are restricted for public-sector use in some jurisdictions. Access to footage is itself access to personal data. Accounts are individual rather than shared, viewing and export are logged, and export requires a stated reason. ## Analytics that reduce work Modern detection distinguishes a person from a vehicle from a branch moving in wind, which is the difference between alerts people act on and alerts people mute. Classic motion detection on an outdoor scene produces so many false alarms that it is usually switched off within a month. We tune object detection, line crossing and loitering rules to the specific scene, and connect them to notifications only where somebody will genuinely respond. Analytics can run on the camera, on the recording server or on a small GPU, and the right answer depends on camera count and whether you want the video to stay on site. ## Who this is for This work suits organisations that already have a system installed and cannot rely on it, sites where surveillance is a regulatory or insurance requirement, and companies whose cameras were installed by a security firm without the IT department being involved and now sit unsegmented on the corporate network. We handle the design, the network, the recording architecture, device hardening and the data protection questions. Where physical installation requires a nationally licensed security company, we work alongside one and remain responsible for the technical design and the systems it runs on. ## Outcomes When an incident occurs, the footage exists, covers the right area, is of usable quality, and can be exported with a record of who exported it. When an auditor asks how long you keep recordings and why, there is a written answer. And the cameras cannot be used as an entry point into the rest of the network. ## What you get - Camera plan with coverage, lens and mounting decisions justified per location - Sized recording architecture with calculated retention rather than an estimate - Dedicated, isolated camera network with PoE budget and switch capacity planned - Hardened devices: no default credentials, no internet exposure, managed firmware - Documented retention, access control and a GDPR record for the footage ## Technologies Milestone XProtect, Genetec Security Center, Network Optix, Frigate, Axis, Hanwha Vision, Bosch, ONVIF ## Frequently asked questions ### How much storage does a camera actually need? It is arithmetic, not guesswork: bitrate multiplied by cameras multiplied by retention. A 4 MP camera at 4 Mbps recording continuously consumes roughly 43 GB per day, so thirty days across twenty such cameras needs about 26 TB before redundancy. H.265 and smart codecs cut this substantially, and recording on motion rather than continuously cuts it further, at the cost of losing the seconds before an event unless pre-buffering is configured. We calculate this per site rather than quoting a round number. ### Can cameras go on the existing office network? They can, and it is one of the most common mistakes we are asked to correct. Cameras are embedded Linux devices that are rarely patched, frequently ship with weak default credentials and have been the basis of very large botnets. They belong on a dedicated VLAN with no route to the internet and no path to the office network except through the recording server. Segmentation also stops continuous video traffic competing with everything else. ### How long may we keep footage? Only as long as the stated purpose genuinely requires. Under GDPR, retention must be justified and documented, and supervisory authorities commonly expect around thirty days for general security monitoring, with longer periods needing a specific reason. Systems configured to keep footage until the disk fills have no defensible retention period at all, which is what turns a routine complaint into a finding. ### Do we need a DPIA? Usually yes. Systematic monitoring of a publicly accessible area is explicitly listed in GDPR as likely to require a Data Protection Impact Assessment, and monitoring employees attracts scrutiny in most member states. The assessment is also the point at which camera angles get questioned, which is generally cheaper before installation than after a complaint. ### Can you integrate cameras with our access control? Yes, and it is where most of the operational value appears. Linking a door event to the footage at that moment turns an investigation from scrubbing hours of video into opening one clip. Both systems should also share the same joiner and leaver process, so that a departure removes the badge and the video system login on the same day.