Vehicle Thermal Camera Resolution Buying Guide

For most commercial fleets, mining operators and specialty-vehicle integrators sourcing an aftermarket thermal camera today, the 384×288 uncooled microbolometer is the pragmatic middle tier: it roughly doubles the pedestrian detection range of a 256×192 unit while costing roughly half the premium of a 640×512 core. Spec 256×192 only when the camera serves as a close-range blind-spot, rear-zone or cabin-facing sensor (this is the tier we ship inside a Consumer Grade Aftermarket mirror unit for passenger cars and SUVs); step up to 640×512 only when the forward-looking channel must resolve a pedestrian or large animal at beyond roughly 250 m on divided highways, or when you are integrating into an Automotive Thermal Camera Modules program that already budgets for long-range L2+/L4 perception. Everything below compares the three tiers on detector pitch, f-number, frame rate, NETD and the real-world conditions under which each tier stops delivering value — including the failure modes buyers do not hear about in vendor demos.

What detection range do I actually get at 256×192, 384×288 and 640×512?

Detection range for a pedestrian (roughly 0.5 m × 1.7 m thermal cross-section) is governed by the Johnson-criteria-style rule of thumb that you need a minimum number of pixels on target to classify "person" rather than "unknown blob". In practice, with a matched f/1.0 lens and a human-sized target:

These numbers are order-of-magnitude, not a datasheet guarantee. Real range collapses quickly when the target is partially occluded by roadside vegetation, wears heavy dark clothing against a cold asphalt background, or when the camera is uncalibrated after a thermal soak in a 70 °C cabin. We at VEHIR publish our on-vehicle pedestrian range under a standardized road test because lab MRTD numbers overstate what a fleet driver actually sees on a wet Tuesday night.

Is 12 µm pixel pitch meaningfully better than 17 µm for a vehicle camera?

Pitch is a parameter buyers confuse with resolution. A 17 µm 384×288 detector has the same pixel count as a 12 µm 384×288 detector, but the larger pixels collect more photons per pixel and deliver better NETD at the same f-number. The trade-off is physical size: a 17 µm 384 array is roughly 30% wider than a 12 µm part, which pushes lens, housing and windshield-bracket design outward.

For an aftermarket consumer-grade mirror unit where packaging is constrained, 12 µm is the only realistic choice. For a commercial vehicle where a 36 × 36 × 46.5 mm core housing is acceptable (the form factor we ship on our professional line), 17 µm can be specified when the operating environment is extremely cold (mining sites, northern-China winter haul roads, Nordic winter fleets) and photon flux is low. If your operating band is −20 °C to +60 °C and most driving is on lit highways, the pitch difference will not be something your drivers report.

When does paying for 640×512 stop paying back?

A 640×512 core is not automatically "safer" — it only pays back when three conditions line up:

1. Road speed justifies the longer range. On a 90 km/h rural road, a 384×288 camera delivering 180 m of detection gives roughly 7 seconds of reaction time, which matches human braking plus steering margin. On a 120 km/h divided highway, that margin shrinks to under 4.5 seconds and the 640 core earns its cost.

2. The display and AI downstream can use the pixels. A 640 stream fed into a 720p mirror display is downscaled on the device; you are paying for resolution the driver never sees. Pair it with a Dual-Light Fusion processor or an MDVR that can ingest the full D1/HD thermal stream.

3. Optics match the array. Cheap plastic-asphere lenses on a 640 core produce soft corners that waste the center resolution. Specifying 640 without a matched germanium or chalcogenide lens is a common procurement mistake.

If your fleet runs at city speeds, on mining sites below 30 km/h, or on campus/closed-loop routes, a 640 core is a spec sheet decoration — a 384 core does the job and leaves budget for more cameras per vehicle.

What NETD, frame rate and spectral band should I actually write into the PO?

Three parameters on the datasheet matter more than resolution for field reliability:

• NETD (thermal sensitivity). Look for ≤ 50 mK at f/1.0, 30 Hz. Below 40 mK the improvement is marginal for driving scenes; above 80 mK, fog and rain scenes turn into noise. We specify our automotive cores at ≤ 50 mK because that is the point where pedestrians in heavy fog remain distinguishable from background clutter.

• Frame rate. 25 Hz / 30 Hz is the minimum for forward driving. 50/60 Hz matters only for high-speed (≥ 120 km/h) highway perception or for any Driving Assistance loop that fuses with radar at 50 Hz. Do not pay for 60 Hz on a rear blind-spot camera.

• Spectral band. Long-wave 8–14 µm is the automotive standard and the only band that performs reliably in fog. Mid-wave 3–5 µm offers better long-range resolution but costs several times more and is almost never justified for aftermarket trucks.

When does a higher-resolution camera itself become a failure risk?

Resolution is not free of failure modes. Procurement teams should flag three:

• Thermal blooming from hot engines and exhaust. A 640 core with a fast lens will saturate on a parked truck's exhaust pipe at close range, washing out nearby pedestrians. This is a tuning problem, not a hardware defect, but it requires your supplier to offer region-of-interest AGC and dead-pixel correction in the field.

• Shutterless vs shutter-based calibration. A shutter-based camera briefly freezes the image every few minutes to recalibrate; that blank frame is unacceptable on a forward ADAS channel. Specify shutterless (mechanical-shutter-free) cores for any forward-facing install.

• Vibration-induced microphonics. Higher-resolution arrays are more sensitive to mount vibration on mining and construction machinery. If you are mounting on a dump truck chassis rather than the cab roof, ask for vibration-tested cores and a damping bracket — this is part of the integration support we offer on the professional line.

Purchasing recommendation: how to pick without over-specifying

For a typical mixed fleet — box trucks, regional long-haul, a small number of mining haulers — we recommend a two-tier spec rather than a single platform:

• Forward long-range channel (highway, rural night): 384×288, 12 µm, 13 mm – 19 mm lens, ≤ 50 mK NETD, shutterless, IP67.

• Side/rear blind-spot and cabin channel: 256×192, 9 mm – 14 mm wide-angle, 25 Hz. This is also the tier we recommend for the Tri-Vision Wide-Angle surround view where per-pixel range matters less than coverage.

Reserve 640×512 for your highway flagship vehicles and for any OEM pre-development program, and validate it on your own routes before fleet rollout — vendor demo footage on dry, clear nights does not reproduce your worst-route rain, fog and deer-crossing season.

Frequently Asked Questions

Can I upgrade a 256×192 camera I already installed to 384×288 later? Mechanically, most of our housings share the same bracket and video output, so a field swap is usually possible. Electronically, the MDVR or mirror display may need a firmware update to accept the higher resolution stream; contact us with your DVR model and we will confirm compatibility before you order.

Does a thermal camera work in a sandstorm or heavy rain the way it works in fog? Long-wave infrared penetrates fog far better than visible light, but heavy rain and blowing sand scatter thermal radiation the same way they scatter radar. Expect range to drop 30–50% in torrential rain; the camera still sees what headlights cannot, but it is not an all-weather guarantee.

What is the typical power draw and operating voltage for an aftermarket vehicle core? Our professional-grade cores draw roughly 1.5–2.5 W at 9–16 V DC, which is compatible with a standard 12 V/24 V truck accessory circuit. Confirm your vehicle's cold-crank voltage dip before tapping the ignition line, since cranking can pull a 12 V bus down to 6 V briefly.

Do I need to certify the camera for E-mark or CE before fleet installation? If you are integrating into a display or DVR that is already E-mark / CE certified, the thermal camera as a sub-component is typically covered; if you are selling a stand-alone windshield-mounted product to a European fleet, you will need ECE R10 EMC on the finished assembly. We provide the component-level test reports integrators need to close their own certification.

How long is the typical field life, and what maintenance does the lens need? The detector has no moving parts (in shutterless configuration) and a rated life beyond 8 years of continuous operation. The germanium lens has a DLC anti-reflection coating and can be cleaned with a standard lens cloth; do not use alcohol, which can cloud the coating over repeated washes.