You walk out of a parking garage into noon sun. Your eyes keep the yellow pillars, the dark car on the left, and the trees on the street. You do not think about it. That is just what seeing feels like.
A camera without enough dynamic range does not do this. It either turns the exit into a white hole, or the garage into a black cave. You can have the street, or you can have the cars. Not both.
That gap, between what you see and what the sensor can hold, is dynamic range. HDR, high dynamic range, is the name we give to the methods that close it.
In one sentence. HDR is the camera keeping both the dark garage and the bright street, the same way your eyes do when you walk into daylight.
What HDR actually is
Dynamic range is the distance between the darkest detail you still need and the brightest highlight you cannot afford to lose. Photographers count it in stops. Each stop is a doubling of light.
A sunny street can easily be twelve to sixteen stops brighter than a corner of a garage. Your eye handles that because the pupil stops down, the retina adapts locally, and the brain builds one picture that feels normal. A silicon photodiode has one well. Fill it, and the street becomes chalk. Underexpose to save the street, and the parked cars disappear.
So HDR is not a filter, and it is not “make the photo look dramatic.” It is extra information: a way for the camera to record a wider stretch of light, then fold that stretch into a picture a screen or an algorithm can use.
If that sounds abstract, use the kitchen window. Sit at the table in the morning. You can still see the fruit bowl. You can still see the garden. Your eye is doing HDR. Point a cheap webcam at the same window and you get a white rectangle and a dark room.
Three scenes you already know
1. The garage exit
This is the cleanest picture of the problem. Inside: concrete, pillars, parked cars in shadow. Outside: sun, sky, a street that is many times brighter. A sensor with a short dynamic range must pick a side.
2. Night driving
Headlights are a small sun. The roadside is almost black. Your eyes still find the pedestrian in a dark jacket. A camera that clips the lamps into a white blob, or crushes the verge to noise, is not “a bit less pretty.” It missed a person.
3. The loading dock
Same physics, industrial clothes. A warehouse is dim on purpose. The yard is daylight. An AMR, a gate camera, or a dock scanner that only sees one of those worlds will stop, ghost, or miss a pallet.
How the real world maps onto the sensor
| What you are looking at | What your eyes do | What a camera without HDR does |
|---|---|---|
| Walking out of a garage | Pillars and street stay | White hole, or a black cave |
| Breakfast by a window | Cup and garden stay | White window, dark table |
| Oncoming headlights | You still see the verge | Lamps bloom; the pedestrian vanishes |
| A dock door at noon | Pallet and truck stay | The yard becomes chalk |
| A doorbell at dusk | Face and sky stay | A silhouette, or a blown sky |
None of these scenes are exotic. They are the default outdoor-to-indoor life of a camera. If your product ever looks through a door, a windscreen, or a window, HDR is not a luxury setting. It is the difference between “the object is there” and “the object is gone.”
Why this matters more for machines than for holiday photos
On a phone, a blown window is ugly. On a robot, it is a missed object. The algorithm cannot classify a truck that has become a white rectangle. It cannot track a person who has become noise. It cannot read a code on a pallet that sat in the crushed shadow.
- Mobile robots and AGVs leave halls into daylight all day. The garage-exit test is their commute.
- Dock and gate cameras sit on the bright/dark border on purpose.
- Dashcams and ADAS live in the headlight problem after sunset.
- Doorbells and building cameras face a bright sky and a shaded porch at the same time.
- Endoscopes and inspection tools leave a cavity into a lit room: the same contrast, smaller hole.
Resolution does not fix this. A 12 MP sensor that clips the highlights still gives you twelve million copies of “no data.” HDR is about whether the data existed in the first place.
How a camera actually does HDR
Ignore the brochure word for a moment. There are a few honest mechanisms, and they are not the same thing.
A bigger native well
Some sensors simply hold more charge before they clip, or read cleaner in the dark. That is native dynamic range. It is the best kind, because you do not have to glue frames together. It is also finite. Physics still wins at a garage door.
Two gains on the same pixel
Dual conversion gain (sometimes called dual analog gain) reads the pixel once as “sensitive” and once as “hard to clip,” then combines the two. Think of it as the pupil having two sizes at the same time. Good for motion, because it is still one exposure.
Several exposures, then a merge
This is the classic HDR: a short shutter for the street, a long shutter for the garage, then an ISP that keeps the good part of each. Sony-style DOL, staggered HDR, and multi-frame phone HDR all live here. It works. It also costs you motion artifacts if the merge is sloppy: a forklift can grow a halo, a car can grow a ghost.
Companding, or a non-linear knee
The sensor squeezes a wide light range into fewer bits before it leaves the chip. Useful when the interface is tight. You must know how the curve was built, or the ISP will guess wrong.
Tone mapping, which is not HDR
Tone mapping is how you show a wide scene on a narrow screen. Done well, it is invisible. Done as “fake HDR,” it is local contrast with a halo around every lamp. It can make a photo look processed. It cannot invent the truck that was never recorded.
A useful test. If the street was a white hole in the raw file, no slider will put the pedestrian back. HDR has to happen at the sensor or in the merge. Makeup comes later.
If you want the module-and-ISP side of this story, why a stacked AI sensor is not the same as a proper HDR pipeline, that is the older Camemaker note on the IMX500 and Flyingchip ISP path.
How to read HDR on a Camemaker sensor page
Do not start from the word “HDR” in a headline. Start from the scene, then look for the mechanism.
- Name the contrast. Garage exit, dock, night road, window. If the product never sees a door to daylight, you may not need the extra cost.
- Look for the method, not the badge. DOL, staggered, dual conversion gain, built-in HDR. Those words tell you how the extra range is made.
- Check the output. Bit depth, MIPI payload, and whether you get a merged HDR frame or two exposures you must fuse on the host.
- Ask about motion. If the scene moves (robots, cars, people), a clean single-exposure or dual-gain path is often kinder than a slow multi-frame merge.
- Keep the lens in the conversation. A lens that flares into the sun will undo a good HDR sensor. Image circle, coatings, and stray light are part of the same picture. That is why the lens library sits next to the sensor library.
We do not invent a stop-count for a SKU that does not publish one. If the datasheet is quiet, treat “HDR” as a question, not a specification.
The engineering conclusion
HDR is not a style. It is whether the camera can do what your eyes already do at a garage door, a kitchen window, a dock, and a dark road.
If your application looks through a bright opening into a darker room, or the other way around, treat HDR as a requirement. Pick the sensor for the method, pick the ISP for the merge, and pick the lens so the sun does not paint a veil over the answer.
That is the work the Camemaker image-sensor library is for: compare the parts on the same terms, then take the garage-exit test seriously.
About Camemaker
Camemaker is the image-sensor and lens library, powered by Camemake, for people who have to match a real scene to a real chip. The Tech Blog is where we write the mechanisms down, not the slogans.