A turnaround changes a site's risk profile overnight. Contractor headcount multiplies, temporary work zones appear where there were none, and the permanent camera plan built for steady-state operations does not reach half of it. By the time crews are into hot-work windows and night shifts, the gap between what your cameras cover and what your site actually looks like is the problem. A full turnaround alert pack should be running, tuned, and trusted by day two. Here is what that looks like and why most sites do not get there.
Why It Breaks Down
Permanent CCTV is built around permanent risk: process units, tank farms, main gates. A TAR adds contractor villages, lay-down yards, scaffold zones, temporary fencing, and vehicle staging areas that did not exist a week earlier and will not exist a week later. None of that is on the fixed camera plan, and running new conduit for a three-week event is not realistic.
Contractor density is the second problem. A unit that runs with forty people on a normal week runs with four hundred during a TAR, many of them from contractor crews who have never worked the site before and do not know where the exclusion zones are. Firewatch and roving patrol were sized for normal operations, not this. Asking the same patrol headcount to cover triple the footprint and population is asking for missed incidents, not preventing them.
Patrol fatigue is real and measurable. A firewatch doing hourly rounds across a sprawling lay-down area at 2 a.m. in August heat is not going to catch a smoldering weld splatter or a contractor who wandered past a barrier tape at hour eleven of a twelve-hour shift the same way they would at hour one. Human attention degrades with fatigue and heat; continuous camera-based detection does not.
The other failure mode is the one that actually kills adoption: alert fatigue from false positives. A video analytics deployment switched on day one with default prompts, cameras not yet matched to zone context, and no tuning window will fire constantly — a shadow misread as a person, a legitimate contractor flagged for incomplete PPE because the prompt has not been adjusted for your site's lighting yet, off-hours rules firing on normal night-shift traffic. After the third or fourth false alarm in a shift, the supervisor stops checking the alert feed. By the time the system might catch something real, nobody is watching it.
- Blind spots by design: temporary zones (lay-down, scaffold, contractor parking) sit outside the permanent camera plan
- Population surge: patrol and firewatch headcount does not scale with contractor density
- Fatigue: night-shift rounds across a larger footprint mean slower response and more misses
- Alert fatigue: untuned prompts generate false positives fast enough that supervisors stop trusting the feed
- Day-one go-live without tuning: treating camera activation as the finish line, not the starting point
The fix is not more cameras or more headcount. It is real-time AI video analytics deployed across the actual TAR footprint, tuned in the first two to three days, and routed so a real alert reaches a person who can act on it — not archived to a recording nobody reviews until after an incident.
What Actually Works
Real-time detection means live camera streams are analyzed continuously on GPU-accelerated infrastructure — a dedicated Vision environment per deployment, not a slow public-cloud API round trip for every check. For time-sensitive alerts like fire and unauthorized zone entry, that processing model matters. Cameras at the turnaround footprint — often brought online through a mobile connectivity deployment covering lay-down areas and scaffold zones — stream to the Vision platform where VLM-based alert templates run against live video.
For oil and gas turnaround deployments, Clover IQ applies a baseline pack plus an industry-specific turnaround pack, auto-provisioned across registered cameras. Below are the five priority risk categories supervisors should expect to be working by day two — not the full list, but the checks that cover the blind spots a TAR creates fastest.
1. PPE compliance — gate, staging, and classified zones
Detects personnel missing required PPE at site entry and in active work zones — hard hat, safety glasses, gloves, and high-visibility or flame-retardant outer garments depending on the area. This is the highest-volume alert category on most TAR sites because of contractor turnover: new crews rotating in daily means a steady trickle of people who have not yet internalized the site's PPE culture. Gate cameras and classified-zone cameras carry different prompt expectations — a contractor village break area should not be evaluated with the same FR coveralls requirement as an active process unit.
2. Perimeter and off-hours intrusion
Flags unauthorized person or vehicle presence in restricted perimeters, no-go zones, and secured areas that should be empty after close. Temporary fence lines around lay-down areas, scaffold zones after the night shift stands down, and contractor parking after gate close are the TAR-specific use cases. Off-hours detection is where shift-pattern tuning matters most: prompts that treat 2 p.m. and 2 a.m. the same will either miss after-hours risk or flag legitimate late-shift activity as an intrusion.
3. Fire, smoke, and open flame
Detects visible flame, smoke, or open flame signatures across hot-work zones, welding areas, and lay-down yards where combustible materials are staged. This is the alert category with the lowest tolerance for delay. It complements, not replaces, your hot-work permit and firewatch program — a continuous check that does not blink. The turnaround pack also includes a separate check for smoking or open flame in hazardous areas where ignition sources are prohibited.
4. Restricted-zone presence
Covers zones where the correct number of people is zero or strictly controlled: hot-work and permit-controlled areas, confined-space entry points, C1D2 classified areas, and other exclusion zones where an unexpected body is itself the alert. These are distinct prompt templates in the turnaround pack — hot-work unauthorized entry, confined-space entry, and classified-area breach — not a single generic motion detector, but the same operational idea: someone is somewhere they should not be.
5. Worker down and environmental hazards
Detects a fallen or collapsed worker on decks, ladders, and turnaround work areas, plus ground-level spill or leak signatures in areas where hydrocarbon or chemical release is a credible risk. Man-down detection is the alert supervisors hope never fires — and the one they want running before hot-work windows open on day one.
The full turnaround pack also includes loitering detection, abandoned-object alerts, camera tamper checks, muster and evacuation headcount support, and gate vehicle logging — provisioned automatically when the alert pack is applied.
- Alert visibility: real-time alerts surface in the Vision dashboard for supervisor review — not buried in a recording log
- Dedicated environment: footage and detection data run in a per-deployment GPU environment under Clover IQ management
- Site-specific tuning: prompt templates are adjusted for what each zone actually looks like during the first 48–72 hours, not left at generic defaults
The First 48–72 Hours
Day-one camera activation is not day-one trust. The gap between "cameras are up" and "alerts are trustworthy" is the tuning window, and skipping it is the single most common reason a video analytics deployment gets ignored by week two.
Baseline false-positive review
For the first shift or two after go-live, every alert gets reviewed against the footage, not just accepted or dismissed at the console. This surfaces the specific triggers your site generates — a particular lighting condition at dusk, a piece of equipment the VLM reads as a person shape from a certain angle, a gate that swings into frame and triggers an off-hours prompt. You cannot tune against a false-positive pattern you have not seen yet.
Camera placement and prompt refinement
Match cameras and prompts to the actual site, not the site plan. Scaffold footprints shift, lay-down areas get reorganized as material arrives, and a camera aimed from the drawing location rarely matches the physical fence line exactly. Expect to adjust camera angles and refine prompt wording in the first two days as the gap between plan and reality becomes visible.
Shift-pattern alignment
Confirm the actual work schedule, not the nominal one. Night-shift crews, off-hours deliveries, and early-morning contractor arrivals all need to be reflected in off-hours and perimeter prompts before intrusion detection stops generating nuisance alerts on legitimate activity.
Prompt and timing adjustments
Each alert template carries prompt wording and timing parameters — how much video context the model evaluates per check. Too aggressive, and you get noise. Too conservative, and you miss developing events. The right settings are site-specific, and they are set by watching real alerts against real footage during the tuning window, not by leaving defaults in place.
Who owns tuning
Tuning is a joint call between HSE and operations, not something left entirely to whoever installed the cameras. HSE owns which zones and alert categories matter for the site's specific risk profile; operations owns what "normal" activity actually looks like shift to shift. Both perspectives are needed to separate a real event from a false positive.
What good looks like by day two versus day five
By end of day two, expect the false-positive rate to have dropped noticeably from the day-one baseline, with the obvious triggers (lighting, equipment shapes, gate swing, shift overlap) already identified and prompt adjustments underway. By day five, the alert feed should be something supervisors actually check without being told to — the signal-to-noise ratio has crossed the point where a fired alert is worth investigating rather than dismissing on reflex. If your system is still generating the same false-positive pattern on day five that it was on day one, tuning did not happen and the deployment needs a hard reset.
The Unit on Your Site
The Clover IQ Mobile Connectivity Unit extends your turnaround footprint with private wireless, push-to-talk, and the camera connectivity to reach lay-down areas, scaffold zones, and contractor villages that permanent CCTV was never built to cover. Video analytics runs on Clover's Vision platform — live streams from the TAR footprint are analyzed with GPU-accelerated VLM inference, and alerts are reviewed in the Vision dashboard. This is the same platform used for turnaround connectivity more broadly, with the alert pack configured to your site's zones and risk profile before the crew arrives.
If you are scoping video coverage for your next turnaround, the conversation worth having is which alert categories in the turnaround pack matter most for your specific footprint, and where your current camera plan actually has gaps. Book a pilot scoping call.
