A typical mid-size manufacturing facility in Southeast Asia carries 40 to 120 CCTV cameras already mounted, cabled, and recording. Before a single AI safety feature goes live, a hardware-first vendor will quote you a partial or complete camera refresh — often 40 to 60 units — without auditing what is already hanging on the walls. That quote can reach $60,000 to $120,000 in camera hardware alone before software, installation, and integration fees are added.
The honest comparison is not "our cameras versus their cameras." It is capex: a new camera stack versus software that auto-recognises the ONVIF cameras your facility already owns. This post works through that comparison in detail — covering ONVIF profiles, bandwidth and retrofit constraints, a total cost of ownership (TCO) table, and a practical readiness checklist — so you can evaluate any safety AI vendor's proposal with clear criteria.
Why camera vendors lead with hardware
Hardware-first vendors have a structural incentive: margin on physical units. A camera refresh is a legitimate sale for a CCTV integrator. The problem is that the pitch rarely starts with an audit of what you have. The question "how many of your installed cameras are ONVIF-compliant?" is almost never the first question asked. It would shrink the quote.
ONVIF (Open Network Video Interface Forum) is a global interoperability standard for IP cameras, maintained by a consortium of manufacturers including Axis, Bosch, Hanwha, Dahua, Hikvision, and Vivotek. ONVIF Profile S covers streaming and PTZ control; Profile G adds on-device recording; Profile T, the most relevant for AI safety applications, adds H.265 support, metadata streaming, and motion region configuration. Most IP cameras manufactured after 2018 support at least Profile S. Cameras from 2015 onward often do as well. If your CCTV estate was installed or refreshed in the last eight years, a meaningful portion of it is almost certainly ONVIF-compliant.
ONVIF auto-discovery uses WS-Discovery — a multicast probe-and-response protocol — so that software can identify, authenticate, and pull video streams from compliant cameras on a local network without manual IP entry or custom driver installation. For a facility running 60 cameras across three buildings, that changes the deployment story significantly. The cameras are already there. The sensor network is already paid for. The remaining question is whether the AI safety software can see it.
What ONVIF auto-recognition actually does
When HyperQ AI Safety connects to a local network, it broadcasts a WS-Discovery probe and collects responses from all ONVIF-compliant devices. For each responding camera, it retrieves the device information, available stream profiles (resolution, codec, frame rate), and capability set. The operator selects which cameras to enrol and assigns detection zones. No custom driver, no proprietary SDK, no camera-brand dependency.
The practical result: a Busan-based manufacturer brought HyperQ AI Safety live on its existing CCTV estate in 1 month from contract to production. No hardware was replaced. The cameras already installed were the sensor network. The month was spent on zone configuration, alert routing, and integration with the facility's shift-management system — not on cabling or camera mounting.
This is the deployment model that hardware-first vendors do not offer, because it eliminates the hardware sale.
The TCO comparison: software-only versus a camera refresh
The numbers below are illustrative order-of-magnitude ranges based on a 50-camera facility, using Hypernology's published pricing and typical integrator quotes for comparable camera hardware. Your facility's numbers will vary by camera age, network infrastructure, and detection-zone count.
| Cost element | New camera bundle (hardware-first path) | Software-only path (existing ONVIF CCTV) |
|---|---|---|
| Camera hardware (50 units) | $25,000–$60,000 | $0 (existing cameras enrolled) |
| Optional thermal cameras (per unit) | $2,250 | $2,250 (add only where needed) |
| Optional IR cameras (per unit) | $650 | $650 (add only where needed) |
| AI safety software licence | $10,000–$15,000 | $10,000–$15,000 |
| Installation and cabling | $8,000–$20,000 | $1,000–$3,000 (configuration only) |
| Smartband add-on (per worker, optional) | $35–$250 | $35–$250 |
| Total estimated first-year cost | $43,000–$95,000+ | $11,000–$18,000 |
| Time to live detection | 8–16 weeks | 2–4 weeks |
The software licence cost is the same in both scenarios. The difference is the hardware and installation lines. A facility with a healthy ONVIF camera estate captures that $32,000–$77,000 gap immediately.
Two caveats worth stating directly:
First, if your existing cameras are below 1080p, positioned poorly for detection zones, or running on a network segment that cannot sustain the additional stream load, some hardware spend may still be justified — but targeted (two or three additional cameras at specific pinch points) rather than a wholesale refresh.
Second, thermal and IR cameras are genuinely different sensor types. If your safety use case requires fire detection in a high-ambient-temperature environment, or fall detection in a low-light area, adding a targeted thermal or IR unit at $2,250 or $650 respectively is rational. The point is that this is a selective decision, not a default.
Detection capabilities once enrolled
Once cameras are enrolled via ONVIF auto-recognition, HyperQ AI Safety runs six core detection classes:
- Fall detection: triggered by rapid postural change and ground-level dwell time, with configurable sensitivity for high-traffic areas
- Fire and smoke detection: frame-by-frame analysis for early ignition signatures, not just billowing smoke
- PPE compliance monitoring: hard hat, high-visibility vest, safety shoe, and glove presence against zone-specific rules — surfaces directly as automated PPE monitoring on ONVIF cameras
- Intrusion detection: configurable zones and time windows, separate from generic motion alerts
- Man-down detection: prolonged motionless posture distinct from seated rest or prone inspection work
- Crowd accumulation: headcount threshold alerts for confined spaces or evacuation lanes
Each class runs on the same video stream. Alerts route to a configurable dashboard, email, or webhook. No separate camera streams are required per detection class.
The smartband add-on extends monitoring to physiological signals — heart rate, blood oxygen saturation, and skin temperature — for workers in high-heat or isolated environments. At $35 per unit entry price, it is an optional layer, not a baseline requirement.
Bandwidth and network considerations before you enrol
This is where software-only deployments require honest scoping. ONVIF auto-recognition finds your cameras, but video analytics software still consumes bandwidth pulling those streams.
Rules of thumb for planning:
- H.264 at 1080p/30fps: approximately 4–8 Mbps per stream
- H.265 at 1080p/30fps: approximately 2–4 Mbps per stream (Profile T cameras)
- Sub-stream processing: most ONVIF cameras expose a secondary sub-stream (typically 640x360 or 720p) at 0.5–1.5 Mbps — AI safety inference can run on the sub-stream and reserve the main stream for recording
- 50 cameras at sub-stream: 25–75 Mbps aggregate — typically within reach of a managed switch with VLAN isolation on the camera network
Most facilities built or upgraded in the last five years run a dedicated camera VLAN at gigabit or better. If yours does not, a network audit is a prerequisite to any camera analytics deployment — including a vendor who is also selling you new cameras.
On-premise edge processing (the server runs in your facility, not in the cloud) keeps video data local. This matters for manufacturers with regulatory or contractual data-residency requirements. HyperQ AI Safety processes on-premise; only alert metadata and thumbnails are logged.
ONVIF estate readiness checklist
Before accepting or rejecting any AI safety proposal, run this checklist against your existing camera estate:
Camera hardware
- Manufacturer and model number recorded for each camera
- ONVIF Profile S, G, or T compliance confirmed (check manufacturer specification sheet or ONVIF conformant products database)
- Minimum resolution 720p per enrolled camera (1080p preferred for PPE detection)
- Sub-stream available at 640x360 or higher
Network infrastructure
- Dedicated camera VLAN or network segment with managed switch
- Available bandwidth 50 Mbps+ on camera segment (for 40–60 camera deployments)
- Static IP or DHCP reservation assigned to each camera
- Network Time Protocol (NTP) synchronisation enabled (required for alert timestamping)
Coverage and positioning
- Entry and exit points for all restricted zones covered by an enrolled camera
- Camera height and angle suitable for fall/PPE detection (generally 3–6 m height, 30–60 degree downward angle)
- No persistent occlusion in detection zones during shift hours
- Lighting adequate for AI inference (minimum 50 lux at floor level for fall detection; thermal or IR recommended below 20 lux)
Access and credentials
- ONVIF device credentials (username/password) available for each camera
- Network access from the AI safety server to each camera IP confirmed (firewall rules checked)
If your estate clears 80% or more of these items, a software-only deployment is almost certainly viable. If it clears 50–80%, targeted additions or camera replacements at specific problem points are likely more cost-effective than a full refresh. Below 50%, a hardware refresh conversation is warranted — but you now have the criteria to scope it precisely.
Where the software path loses
Directness requires naming the scenarios where software-only is not the right answer.
Cameras older than 2012: Pre-ONVIF analogue cameras, or early IP cameras running proprietary protocols, cannot be enrolled. If more than 30% of your estate falls into this category, the economic argument for selective hardware replacement strengthens.
Coverage gaps in critical zones: If a high-risk confined space or chemical handling area has no camera coverage at all, adding a camera is the right answer. The decision is "add one camera where needed" not "replace everything."
Sub-1080p cameras in PPE detection zones: PPE compliance monitoring at distances over 8 metres requires at minimum 1080p resolution. A 720p camera at 12 metres will produce unreliable hard-hat detection results. Targeted camera upgrade at that zone is justified.
Network segments with no available bandwidth: A camera network running at sustained 90%+ utilisation cannot absorb additional stream pulls without degrading recording quality. Network remediation or edge recording adjustment is a prerequisite.
None of these scenarios require a 50-camera refresh. They require a targeted assessment. A vendor who quotes you a full camera refresh without conducting this assessment first is selling hardware.
ISO 45001 and the evidence trail
AI safety monitoring contributes directly to ISO 45001 compliance documentation — specifically clause 9.1 (performance evaluation) and clause 10.1 (incident investigation). Automated detection logs provide timestamped, camera-attributed evidence of PPE compliance rates, intrusion events, and fall incidents. This is more auditable than manual supervisor rounds.
The distinction between a hardware-dependent and a software-based deployment does not change the compliance evidence structure. What changes is whether ISO 45001 evidence from existing cameras is accessible from day one on your current estate, or whether it waits behind a hardware procurement cycle of 8–16 weeks.
For facilities already under an ISO 45001 audit cycle, the 2–4 week deployment timeline on existing cameras versus a 3–4 month hardware refresh timeline is not a marginal difference.
Evaluating any vendor's proposal
When a safety AI vendor presents a proposal, five questions sharpen the comparison:
Did the proposal begin with an ONVIF audit of our existing cameras? If not, the pricing baseline is not anchored to your actual situation.
What percentage of the camera line items are replacements for existing ONVIF-compliant units? Each replacement camera that substitutes for a compliant existing camera is hardware cost with no detection capability gain.
Is the AI safety software portable to cameras we procure separately? A software licence that works only with the vendor's own camera SKUs is a hardware-locked architecture, regardless of how it is described.
What is the sub-stream processing capability? If the system requires full main-stream access from every enrolled camera, bandwidth and storage costs increase substantially.
Where does video data reside? Cloud-based inference requires a permanent data path outside your facility. On-premise processing does not.
A vendor with a strong software answer gives direct responses to all five. One selling a camera-dependent architecture will either avoid the questions or reframe them as technical concerns that require "proper integration."
The decision point
The question is not whether new cameras are ever justified. Sometimes they are. The question is whether the default assumption of a full camera refresh reflects your facility's actual state, or whether it reflects the vendor's revenue model.
1 month to live AI safety monitoring on an existing CCTV estate at $10,000 in software versus $43,000–$95,000 for a hardware-bundled deployment: the TCO gap is large enough that it deserves a clear answer grounded in your specific camera inventory, not a quote generated without one.
ONVIF auto-recognition exists precisely to make that camera inventory visible before you spend anything. The Busan-based manufacturer's experience is instructive not because 1 month is a universal deployment timeline — your facility's network complexity and zone count will determine that — but because the deployment timeline was set by configuration work, not by hardware procurement. That distinction is available to any facility whose CCTV estate meets the readiness criteria above.
The vendors who will not do an ONVIF audit before quoting are telling you something about their business model. The ones who start with the audit are telling you something about their confidence in the software.
Before accepting a camera-refresh quote: send us your existing camera inventory (make, model, installation year) for 80 cameras or fewer, and we will return an ONVIF compatibility assessment and a software-only deployment cost estimate within 5 business days. No contract is required, and the assessment is yours to use regardless of what you decide. Request your ONVIF compatibility assessment