TopeLeave about Dec 1
← Journal Draft · not published

Cameras, inside and out

2026-08 Draft
Whole box Systems

This started as a question about the drone-style overhead view and ended somewhere more useful. There are two real jobs here: seeing enough to place 24 ft of box on a bad track, and watching the whole perimeter while it is parked somewhere else. They turn out to want almost the same cameras, and the cheapest way to get both is also the way that keeps the software ours.

What is already fitted

Two rear-facing cameras, and they are close to redundant with each other. One is the factory camera, which shows on the dash. The other is a Wolfbox G900 TriPro, a three channel mirror system with a 4K windshield camera, a front bumper camera and a rear camera, on a 12 in mirror display. Its real advantage over the factory one is that it is always on while driving rather than only in reverse.

Both rear cameras are mounted high. That is the problem.

Why height is the problem

A camera high on the back of a tall box sees a long way down the road and almost nothing near the bumper. The ground close behind the van falls under the lens, and that strip is exactly where a rock, a post or a picnic table lives. Two cameras in the same high position do not fix each other. What fixes it is a second camera low down, which sees the near ground and very little else. Box trucks run both for this reason.

WHY A LOW REAR CAMERA BOX high camera · fitted twice already low camera · the gap low camera covers it DEAD ZONE the close ground behind the box, invisible from up high solid = high camera · dashed = low camera Schematic. Angles indicative, no position measured.

The corners, and the whole perimeter

The other thing worth seeing is the corners. Placing 7 ft of width and 14 ft of box on a rocky track is a corner problem, not a rear problem, and neither existing camera helps. Separately, the van needs to be watchable while parked somewhere else, and that is a perimeter problem: not one view, but no blind side.

Those two requirements mostly collapse into one set of hardware, but not in the obvious way. The instinct is to put a camera on each corner. Drawing it shows why that fails: a camera at a corner aims diagonally away from the van, so it sweeps the diagonal and leaves the middle of every flank uncovered. Corners are where cameras look from worst, not best.

What actually closes a perimeter is one camera at the middle of each face, aimed straight out. Each covers its own side, and adjacent cones overlap past the corners. That is exactly where the 360 kits put their four cameras, and this is the reason. The corner then gets seen by two cameras from two angles, which is better for judging a corner than one camera sitting on it.

PERIMETER FROM FOUR FACES BOX 14 ft × 7 ft cab end rear one camera at the middle of each face, aimed straight out, cones overlapping past the corners the wedges here are 140 degrees, so the small unlit slivers at each corner are real, not drafting Schematic. Box to scale, camera positions proposed.

What four cameras do not solve

They are mounted high, so they inherit the same near-ground blindness as the existing rear cameras, just at four places instead of one. For security that is fine and arguably correct: what matters there is seeing somebody approach, which happens further out. For driving it means the low rear camera is still a separate item and not something a perimeter camera absorbs.

The corner slivers in the drawing are real. Closing them completely would take a true fisheye at 180 degrees or more, which is what the 360 kits fit and why their spec sheets quote 190 and 230 degrees. That is a genuine trade: the wider the lens, the thinner the pixels are spread, and identifying somebody at 20 or 30 ft is the thing a security camera is for. A gap two feet from the skin at the corner matters much less than being able to tell who is standing at the door, so the moderate lens is probably right and the gaps are worth accepting knowingly.

Same trade-off decides how good a future software stitch could be. A bird's eye built from four 140 degree cameras would have those corner gaps in it. One built from four fisheyes would not. Worth knowing before ordering lenses, not after.

Why not a 360 kit

The overhead bird's eye is real, and one vendor literally sells it as a helicopter view: four fisheye cameras and an ECU that stitches them into a synthetic overhead with a model of the van in the middle. Three reasons it is the wrong buy here.

1

It is a ground-plane illusion

The stitch assumes everything it sees is flat pavement. Anything with height leans away from the van and smears at the seams. It answers "where is my corner relative to that rut" well and "how tall is that rock" not at all. Useful at walking pace, a novelty at speed, because everything in the frame that matters is within 10 to 20 ft of the van.

2

Roughly $1,500 buys the stitch, not the cameras

The kits run from about $200 for generic AHD hardware up to $1,236 for the RVS inView, $1,600 for the Brandmotion SurroundVUE, $1,851 for the Rosco and about $3,060 for the Brigade. Four separate views on one screen cost a fraction and lose nothing except the overhead illusion.

3

The ECU is a sealed box doing the interesting part

The standing principle on this build is to write the software rather than depend on somebody else's. A 360 ECU is exactly the shape of thing that principle warns against: the stitching, the views and the triggers all live somewhere we cannot reach. Four open cameras keep that door open, because the stitch can be done in software later. Buying the ECU closes it.

One thing that got over-flagged in the first draft of this page and is worth correcting: the box is 7 ft wide against a cab of about 6 ft 8, so the box stands roughly 2 in proud per side. That is not the awkward shape problem it was made out to be, and it is not a reason to spend Brigade money.

One set of cameras, two jobs

The first version of this page argued for keeping driving cameras and security cameras rigidly apart, on the grounds that a reversing view must never depend on the house system. The rule is right. It is also already satisfied, by hardware that is fitted: the factory camera and the Wolfbox are both low-latency, always on, and independent of everything else in the van.

Which means the new cameras can be network cameras without a safety problem. They are a placement aid at crawl speed, not the primary reversing view. At 2 mph, 200 ms of network latency is about 7 in of travel, which is fine for judging a corner and would not be fine as the only thing you were reversing on.

And they earn their keep twice. The same four cameras that show the corners while manoeuvring are the ones holding the perimeter while the van is parked somewhere else. That is the whole argument for buying network cameras rather than a driving aid: a 360 kit would have given the overhead picture and then been useless the moment the ignition went off.

Live view depends onThe router, not the Mac mini. The camera serves its own stream, so a phone on the van WiFi works with everything else off
RecordingCamera-side microSD, so clips survive the Mac mini being down
DetectionFrigate on the Mac mini, alerts out over cellular
Away linkStarlink Mini, with the cellular SIM as backup. Both stay up, so this is full bandwidth rather than compressed alerts

Hardware, and an honest caveat

12V native rather than PoE. Reolink cameras are 48V PoE only, so a 12V bus would need a boost injector or a 12V-input switch. Amcrest and Dahua generally put a 12V barrel on the pigtail as an alternate input, so they run straight off the house bus with no conversion stage at all. Candidate is the Amcrest IP8M turret family, IP67, microSD to 256 GB, RTSP and ONVIF, roughly $80 to $130 each.

The caveat: consumer cameras are rated to about 60°C and are not vibration rated. Baja sun and washboard will kill them eventually. Automotive and marine cameras that survive both run $300 to $600 each, which does blow the doors off. So the plan is to treat them as consumables. They are cheap, they are replaceable, and the part that is genuinely hard to change later is the software architecture, not the camera hanging off the corner.

Whether leaving them on is affordable

Comfortably, yes.

CamerasFive at 5 to 12 W each, so roughly 25 to 60 W
Starlink MiniAbout 15 W idle, 20 to 25 W steady. Not the 50 to 75 W of the standard dish
Whole lotCameras plus dish plus router is on the order of 70 W, about 1.7 kWh a day
Reserve810 Ah at 12 V is roughly 10 kWh, so around six days with no solar at all
SolarAbout 1400 W over six panels, which covers it several times over in Baja

This corrected a live design decision rather than just a number. The electrical notes had AWAY mode running on cellular with the dish woken on demand, purely because Starlink was recorded at standard-dish power. At Mini figures it can simply stay up, which means full bandwidth to the cameras from anywhere instead of small compressed alerts, and the SIM goes back to being the backup for when the dish is blocked.

Storage is the part to be careful with. Four cameras at 1080p H.265 recording continuously is roughly 2 TB for 30 days. In a vehicle that means SSD, not a surveillance hard drive, because of vibration. Better still, do not record continuously and let Frigate keep events only.

Inside

The dog camera has a clear job and ties into the dog-safety node already planned: mmWave presence, temperature, and control of the air conditioning. The camera is what lets you confirm with your own eyes what the sensors claim, which is the entire point when the failure mode is a dead dog.

Interior cameras go on a physically separate switched circuit from the exterior ones. The control panel already hard-cuts camera power with a rocker; this splits it in two. That does not stop the interior cameras being live while away, which is wanted. It means they can be genuinely dead, not software-dead, when the van is lived in.

Interior audio recording defaults off. It is the least useful footage and the murkiest legally. Vehicle cameras are broadly fine in Mexico and recording inside your own vehicle is generally permitted, with state-level variation, but nothing here should depend on that answer.

Mounting, and what the shell actually is

The body is a one-piece molded fiberglass Unicell shell reinforced with an internal 8020 aluminum frame. Worth stating plainly because the notes previously called it an aluminum body: the shell is RF-transparent and the 8020 is a lattice rather than an enclosure, so it is not a Faraday cage and WiFi inside the box works. That has now been corrected on the systems page.

For mounting, the frame is good news. The roof rack is already bolted through the roof on 1020 feet into an internal 1020 frame, which is a proven detail on this vehicle for hanging load off the shell, and camera mounts should follow it rather than invent something. Where the 8020 runs, bolt to the frame. Where it does not, bond and mechanically retain, and remember that adhesive alone loses strength above 60 to 70°C, which a roof in Baja sun exceeds.

What is still open

Needs Jeff
  • Where the 8020 runs relative to the middle of each face and the bumper line. This is the only thing blocking a mounting detail.
  • Whether the low rear camera mounts to the box or the chassis. Lower is better for the near ground, but it is more exposed to spray and stone strike and it crosses the box-to-chassis boundary for cabling.
  • How the corner views get displayed while driving: a tablet on the van WiFi, a second small monitor, or $555 of Naviks interface onto the SYNC 4 screen. The Naviks path takes AHD rather than IP, which would split the camera set in two, so probably not.
  • Lens angle. Roughly 120 degrees favours identifying somebody at 20 or 30 ft; a true fisheye favours a future stitch. They pull against each other and security should probably win.
  • Whether the interior cameras count toward the same five or sit on their own budget.

The cameras, named

Tags follow the same convention as the water system: the letter is the group and the number is the position within it. V is driving vision, S is perimeter security, I is interior, N is network and power, M is a monitor. The schematic and the table below use the same tags, so neither can quietly drift from the other.

CAMERA SCHEDULE AND LINKS wired remote DRIVING VISION V1 · factory rear high · dash screen · reverse only V2 · Wolfbox rear high · 50 ft cable · always on V3 · Wolfbox front 4K windshield V4 · Wolfbox bumper low front · already fitted V5 · low rear NEW · covers the dead zone PERIMETER · SECURITY S1 · perimeter front mid-face · IP S2 · perimeter driver mid-face · IP S3 · perimeter rear mid-face · IP S4 · perimeter pass. mid-face · IP INTERIOR I1 · cabin dog-safety node · privacy circuit M2 · dash screen SYNC 4 · factory M1 · Wolfbox mirror 12 in · always on while driving N2 · 12V bus + switch camera power and network N4 · Starlink Mini 15W idle · 20-25W steady N1 · router GL-X3000 dual WAN failover N5 · cellular SIM Telcel + US · backup N3 · Mac mini Frigate · event clips M3 · you, anywhere live view + alerts remote, via N4 or N5 V = driving vision · S = perimeter security · I = interior · N = network · M = monitor V1 to V4 are fitted today. V5, S1 to S4 and I1 are proposed. Schematic. Links are logical, not a wiring diagram.
TagWhatWhereStatusNotes
V1Factory rearHigh on the box, rearFittedShows on the dash. Reverse only
V2Wolfbox rearHigh on the box, rearFitted50 ft cable. Always on while driving
V3Wolfbox frontWindshieldFitted4K
V4Wolfbox bumperFront bumperFittedThe low front angle, already covered
V5Low rearRear, bumper heightProposeThe dead zone no high camera can see
S1Perimeter frontMiddle of the front faceProposeIP, 12V native
S2Perimeter driverMiddle of the driver flankProposeIP, 12V native
S3Perimeter rearMiddle of the rear faceProposeIP, 12V native
S4Perimeter passengerMiddle of the passenger flankProposeIP, 12V native
I1CabinInteriorProposeDog-safety node. Own privacy circuit
M1Wolfbox mirrorCabFitted12 in. Takes V2, V3, V4
M2Dash screenCabFittedSYNC 4. Takes V1
M3You, anywhereRemoten/aLive view and alerts over N4 or N5
N1Router GL-X3000Electronics bayOwnedDual WAN failover
N212V bus + switchElectronics bayOn BOMCamera power and network. Port count follows camera count
N3Mac miniElectronics bayPlannedFrigate, event clips
N4Starlink MiniRoofOwned15W idle, 20 to 25W steady
N5Cellular SIMIn N1OwnedTelcel plus a US carrier. Backup WAN

Five cameras are proposed: V5 plus S1 to S4, with I1 inside. Nothing new is bought and no position is fixed. Full research notes, the 360 price comparison and sources are in planning/cameras.md.