The cameras are running. The DVR is full. The trailer monitor is on. And nobody is watching. That is the entire construction site security problem in one sentence.
Construction is the trade where camera placement matters most and changes most. The lay-down area moves with every phase. The building grows and blocks shots that were good in week one. The recorder is fine; the DVR does its job. The piece that is broken is the part nobody sells: the layer that watches the feed and tells the project super at 2am that someone is on the materials pile. This page covers that gap and the edge AI overlay that closes it.
See an overlay reading a real construction DVRThe shape of the gap
The gap is not a missing camera. It is not a missing recorder. It is not even a missing network. It is the absence of a person, or a process, or a piece of software, that looks at the live feed and decides whether a human is doing something that should generate an alert. On 95 percent of active job sites that role is unfilled.
The trailer wall monitor exists. It shows the grid. The estimator and the project engineer glance at it during the day. After 6:30pm the trailer is locked and nobody glances at anything. Between 6:30pm and 6:30am, every shift is staffed by a recording device, not a watcher. The DVR writes frames to disk and the disk wraps every two to four weeks.
The forensic value of that recording is real. The deterrent value, in practice, is not. A thief who has done it before knows the camera is recording but not watching. The cycle (steal, get away clean, get a clip pulled the next morning, walk on the police report) is well understood on the supply side. Closing the gap means making the cameras feel like they are watched, in real time, all the time.
Why placement is the differentiator on construction sites
On a finished property the building is a fixed obstacle. On a construction site the building is a growing obstacle. Cameras that had a good view in March have an obstructed view in May because the third floor went up. The pole that was in the middle of the lay-down is now in the middle of nothing because the lay-down moved 200 feet north.
Why phase one camera placement does not survive phase three
Phase one (excavation)
Cameras placed on conex corners and the trailer. Good sight lines to perimeter and lay-down. Working as designed.
Phase two (foundation and slab)
Lay-down moves. Two cameras now look at empty asphalt. One camera is lifted by a sub to make room for forms.
Phase three (framing and shell)
Building itself blocks the most useful angles. New gear arrives and stages where the cameras cannot see it.
Phase four (finishes)
Half the cameras are pointed at walls. The remaining half watch what is left of the lay-down, which is mostly small tools.
The overlay that reads the trailer DVR
An edge AI overlay does not solve the placement problem; that is still on the field team. What it solves is the watching problem. One small fanless box sits next to the DVR, takes the HDMI feed from the recorder, runs a model on each tile in the grid, and fires alerts when a person or vehicle hits a defined zone during an armed window. The cameras themselves never know there is an overlay; they keep doing what they were doing.
The four loops of an overlay
Capture
HDMI from the DVR feeds the overlay's capture card. Frames at the recorder's native rate, downsampled to whatever the model needs.
Layout learn
Detect 4x4 / 3x3 / 2x2 grid; bind each tile to the DVR's text name strip; remap if the operator changes the layout.
Detect and filter
Person and vehicle detection per tile; per-zone gates; per-camera arming windows; per-event dwell thresholds.
- 4
Alert and persist
WhatsApp or SMS to the project super; append the event to a local file with a monotonic counter so nothing is lost during a 4G outage.
Active monitoring options for a job site
Worth being honest: an overlay is one of three reasonable ways to close the gap. The right choice depends on the site, the budget, and the tolerance for false positives.
| Feature | Other ways to close the gap | Edge AI overlay (Cyrano style) |
|---|---|---|
| Always-on review | Remote monitoring service: human in the loop, costs 800 to 2,000 / month | Software runs 24x7, no per-hour cost |
| Response time | 60 to 300 seconds (depends on staffing model) | 5 to 15 seconds from event to WhatsApp |
| Setup | Forward RTSP for each camera, sometimes per-stream license | One HDMI cable, no DVR config changes |
| Behavior on 4G outage | Remote service goes blind | Detection continues, queued in local outbox |
| False positive handling | Human filter (good at context, slow) | Zone + dwell + arming filter (good at speed, no context) |
| Best together | Service-only is expensive and slow | Overlay does triage, escalate to human service for high-confidence flags |
What the overlay covers, and what it does not
Honesty matters here. An overlay is a watching layer; it does not improve the optics, the placement, or the lighting. The cameras have to see the relevant scene for the overlay to fire on the right things. If the camera is pointed at a wall, the overlay is pointed at a wall.
overlay covers:
- Person and vehicle detection on every tile
- Per-camera arming schedules (workday vs after-hours)
- Per-zone rules (lay-down, gate, fuel tank, trailer door)
- Dwell-based alerts (someone standing at the gate for 12 seconds)
- WhatsApp delivery with thumbnail
- Local outbox so events survive a carrier outage
overlay does not cover:
- Bad camera aim or obstructed view
- Dim lighting at night without IR-capable cameras
- Theft inside the building shell where there are no cameras
- Identity recognition of the person in the frame
“On one Class C multifamily property in Fort Worth, simply having an overlay watch the existing DVR for 30 days surfaced 20 incidents that the recorder had captured but nobody had reviewed, including an attempted break-in. Same cameras, same wiring. The only thing that changed was that the feed had a watcher.”
Cyrano field notes, multifamily deployment
How a project superintendent actually uses it
The point of the overlay is not the dashboard. The point is the phone in the project superintendent's pocket. WhatsApp pings at 2:14am with a thumbnail and a one line caption: gate-main, 12s dwell, vehicle present. The super calls the on-call subcontractor or dispatches a driveby. By the time anyone arrives, the truck has either left or been intercepted, and the next morning the crew shows up to a site that is exactly the way they left it.
That is the loop. There is no log review. There is no morning footage pull. There is no police report after the fact. The cost of the overlay is the small monthly software fee plus the one time hardware. The benefit is that the loop closes in seconds instead of hours.
The piece nobody was selling
The DVR is for evidence. The overlay is for prevention.
Recorders were sold for 25 years as security cameras when they were really security recorders. The piece that does actual security (the watcher in the loop, who turns a recording into a phone call) was always either a guard or nobody. On a moving construction site, the guard is expensive and the nobody is the default. An overlay is the third option: a small piece of software, near the DVR, that makes the cameras feel watched all the time.
See an overlay running on a real construction DVR
A 15 minute call. We connect to a live job-site DVR, draw zones on three cameras, and fire a test WhatsApp alert while you watch.
Construction site DVR monitoring gap: frequently asked questions
What does the phrase 'we had cameras but nobody was watching' actually describe on a construction site?
It describes a recording system that is doing exactly what it was sold to do (writing 24x7 frames to a hard drive in the trailer) and nothing else. After a theft, the GC pulls the recorder, finds the relevant clip, hands it to the police, and accepts the loss. The cameras did their forensic job. They did not deter, did not interrupt, and did not alert. That gap (between recording and watching) is what an active monitoring layer fills.
Why is camera placement on a construction site different from camera placement on a finished property?
Because the site changes shape every two to four weeks. Phase one is excavation and the cameras need to see the perimeter and the materials lay-down. Phase two is foundation and the lay-down moves to a different corner of the lot. Phase three is framing and the building itself blocks the view from the original camera positions. By the time the project is in finishes, half the cameras are pointed at the wrong place. Permanent-property camera planning assumes the building is the building. Construction-site camera planning has to assume the building is a moving target.
What does an edge AI overlay actually do that closes the monitoring gap?
Three things. First, it reads the DVR HDMI feed (the same image the trailer wall monitor shows) so it does not require the cameras themselves to be replaced. Second, it runs detection on every tile in the grid, applies per-camera and per-zone rules (after-hours arming, dwell thresholds), and fires WhatsApp or SMS alerts when those rules trigger. Third, it persists every event to a local file so the alerts survive the carrier outages that 4G hotspots routinely produce. Result: the cameras keep recording, the DVR keeps doing its forensic job, and the overlay does the watching that nobody was doing before.
How is reading the HDMI feed different from talking to each camera over RTSP?
RTSP requires per-camera credentials, an open port for each stream, and often a license tier on the recorder. On a typical site DVR those are either disabled, undocumented, or behind a paywall. HDMI is the universal escape hatch: every recorder has a video output, the trailer monitor is already plugged into it, and the overlay box can sit in the loop. The overlay learns the active grid layout, binds each tile to a camera name from the DVR's name strip, and treats the grid as N independent video sources. Setup time is one cable; no DVR configuration changes are required.
What is the difference between a person walking past at 6pm and a person walking past at 2am, from the overlay's perspective?
The arming window. Each camera has a per-day-of-week schedule that says 'detection is suppressed during these hours' (typically 6:30am to 6:30pm during the working week) and 'detection fires alerts during these hours' (typically 6:30pm to 6:30am). The 6pm walker hits the suppressed window and never generates an alert. The 2am walker hits the armed window, the dwell counter starts, and once the threshold is crossed the WhatsApp message fires. The same person, same zone, same camera, totally different behavior depending on the schedule.
Why is the overlay better than hiring a remote monitoring service to watch the DVR feed?
Both approaches cover the gap. A monitoring service is staffed by humans who can read context, but it costs 800 to 2,000 dollars a month per site and the response time is the staffing model. The overlay is software that runs all the time, costs less, and produces an alert in the inbox of the person who can actually act (the project superintendent), in seconds. The honest answer is hybrid: overlay does the always-on detection and triage; the monitoring service is escalated to only when the overlay hits a confidence threshold the GC defined.
Comments (••)
Leave a comment to see what others are saying.Public and anonymous. No signup.