C-DRONE GUIDE · 31 AUGUST 2026

Drone Sensor and Shutter: Global vs Rolling Shutter, and What It Changes on a Survey

« We need a global shutter. » The phrase travels through drone survey tenders like a password, copied from one specification to the next without anyone knowing what it costs or when it is warranted. Sometimes it is : a sensor reading its image line by line while the aircraft moves at 8 m/s geometrically distorts every frame, and that distortion propagates into the bundle adjustment. Sometimes it is not : on slow flight, in stop-and-go, on a visual inspection, it adds nothing and raises the invoice. In between sit some simple physics, published figures, a correction model in photogrammetry software — and the assumptions under which that model holds. This guide sorts it out, for the shutter as well as for sensor size and lens, and ends with what a client should write in the tender and read in the processing report.

Published on 31 August 2026, reviewed on 11 September 2026 — regulations in force as of September 2026.

How a sensor reads an image: line scan, readout time, global shutter

A digital camera sensor does not « take » an image in one block : it exposes it and then reads it, and the way it reads it separates the two families. In a rolling shutter — near-universal on compact CMOS sensors — rows of photosites are exposed and read one after another, from the top of the frame to the bottom. The top of the image is therefore not captured at the same instant as the bottom : between the two elapses the sensor's readout time, a few tens of milliseconds depending on the model. In a global shutter, every row is exposed simultaneously and then transferred : the whole image corresponds to a single instant. A mechanical shutter achieves the same result by another route : a physical part closes off the sensor, readout then happens shielded from light, and the timing offset between rows disappears from the image.

As long as the camera does not move, the difference is invisible. As soon as it moves, it becomes geometry. While the sensor scans, the drone keeps flying : the bottom row of the frame is taken from a slightly different position than the top one, and the image is sheared. The scale of the effect is no mystery : displacement in pixels = readout time × ground speed ÷ ground resolution. Readout times published by software vendor Pix4D for consumer drone cameras give the order of magnitude : 74 ms on the DJI Phantom 2, 33 ms on the Phantom 3 and 4, 30 ms on the Inspire 1 (FC330) and on a GoPro Hero 4 Black.

The worked example published by the same vendor is telling : a Phantom 2 Vision+ flying at 8 m/s, 70 m above ground, travels close to 60 centimetres during sensor readout alone, which translates into roughly ten pixels of vertical displacement in the image. Ten pixels, on a frame meant to constrain a model claimed accurate to the centimetre, is not an engineer's subtlety : it is a systematic error, oriented along the flight direction, that propagates into the bundle adjustment. And it is proportional to speed : the same camera hovering produces no such distortion at all.

One point of vocabulary, because it causes misunderstandings in tenders : on mapping drones actually on the market, what is offered is almost never an electronic global shutter but a mechanical shutter. The DJI Phantom 4 RTK (1-inch sensor, 20 MP), the Mavic 3 Enterprise (4/3 sensor, 20 MP) and the Zenmuse P1 payload (full frame 35.9 × 24 mm, 45 MP) all three carry a mechanical shutter alongside their electronic one. Writing « global shutter » into a specification therefore disqualifies, to the letter, nearly every mapping drone genuinely available. The right wording targets the effect sought, not the technology : no rolling distortion at the chosen flight speed.

What rolling distortion really costs on a survey

The useful question is not « does it distort? » but « by how much, and does that blow my error budget? ». Two published works answer with figures rather than impressions.

The first is the reference study on the subject : J. Vautherin, S. Rutishauser, K. Schneider-Zapp, H. F. Choi, V. Chovancova, A. Glass and C. Strecha, « Photogrammetric Accuracy and Modeling of Rolling Shutter Cameras », published in 2016 in the ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences (DOI 10.5194/isprs-annals-III-3-139-2016). The authors compared several datasets acquired on grid flight plans with consumer rolling shutter quadrotors, and one acquired with a mapping fixed-wing fitted with a global shutter camera. Three lessons emerge. First, adding a rolling shutter model to the bundle adjustment brings a significant accuracy improvement on validation control points for faster flights (8 m/s). Second — and this is the sentence tenders forget — competitive accuracies are reachable with the correction model, yet global shutter cameras remain superior. Third, the authors show that drone speed and direction can be estimated from the rolling shutter effect alone, which says a great deal about how structured, and how far from random, that effect is.

The second gives the order of magnitude in centimetres, on a field case many engineering firms will recognise. A. H. İncekara and D. Z. Seker published in 2021, in the International Journal of Environment and Geoinformatics, a study on the rolling shutter effect on the accuracy of a photogrammetric product from a low-cost UAV (DOI 10.30897/ijegeo.948676). Two flights over roughly 60 hectares with a DJI Phantom 4 Pro, at 8 m/s then 12 m/s, processed twice — with and without correction — and checked against 24 ground control points. Result : total root mean square error falls from 6.33 to 4.78 cm for the 8 m/s flight, and from 7.01 to 4.00 cm for the 12 m/s flight.

Translated for a buyer. On a survey whose contractual requirement sits around 5 cm — a condition record, earthworks monitoring, a development orthophoto — the margin is comfortable once correction is applied : that is not where the project is decided. On a survey where the requirement drops to one or two centimetres — setting out, as-built compliance checks, comparing two campaigns to detect movement — two to three centimetres of systematic error eat most of the error budget before georeferencing, control points or matching quality have even been discussed. That is exactly the reasoning we set out in our guide to RTK, PPK and when to actually require them : the accuracy of a deliverable is not granted by a component, it is the outcome of a chain in which every link consumes its share.

Software correction: what it recovers, and under which assumptions

Professional photogrammetry software can model rolling shutter. The principle is simple : instead of assigning a single position and orientation to the camera for the whole frame, the bundle adjustment linearly interpolates between the position at the start of readout and the position at the end, and assigns each pixel row the corresponding position. The distortion stops being an error and becomes a model parameter. This is the approach described by Pix4D, which recommends enabling the model as soon as the computed vertical displacement exceeds 2 pixels — a threshold easy to check beforehand, since it depends only on camera readout time, programmed ground speed and target ground resolution. An equivalent exists in the other suites on the market, labelled « rolling shutter compensation », to be enabled in the camera calibration settings.

What matters most is what product sheets leave unsaid : linear interpolation assumes linear motion. The model is at its best where motion is steady and orientation stable — the vendor itself notes the method works especially well with multirotors whose gimbal keeps the camera facing nadir. Every departure from that assumption erodes its effectiveness :

• In turns and at the end of strips, the path is not straight and the gimbal is catching up ; frames taken during those phases are the least well corrected. Good reason not to trigger in turns, and to be wary of overly tight flight plans.
• In gusts and turbulence, the aircraft undergoes rapid accelerations during the few tens of milliseconds of readout. Motion is no longer linear, it is erratic : the model cannot represent it.
• On façade flights, camera horizontal and drone translating sideways, rolling shutter combines with an oblique geometry in which scale varies strongly across the frame : the least favourable configuration, and precisely the one used for a façade 3D model from oblique photogrammetry.
• On close-range structure flights, ground resolution becomes very fine — a centimetre or less per pixel — and the same metric displacement translates into far more pixels. The 2-pixel threshold is crossed at ridiculously low speeds.

The practical conclusion is clear. Software correction is an effective safety net for a nadir grid survey, at controlled speed, in calm weather ; it is not a licence to fly fast, nor a substitute for hardware when the geometry leaves that envelope. And it always leaves a residual : the most robust lesson from the 2016 Vautherin study, which concludes that global shutter cameras remain superior even after correction.

When a mechanical shutter is essential — and when it changes nothing

Asking the question this way round saves a lot of money. A mechanical shutter is not a general quality tick-box : it answers one specific constraint, motion during acquisition. It becomes decisive in four situations.

A large area to cover in a short window. A hundred hectares, one morning, shifting light : you do not cover that flying slowly. You need to hold 10 to 15 m/s, and at that speed rolling distortion is no longer a residual. This is the classic quarry, logistics platform or development zone survey — the use case covered by our drone surveying and photogrammetry page.

A contractual centimetre requirement. As soon as the deliverable commits you — setting out, as-built compliance checks, comparison of two campaigns months apart — every error contribution must be defensible before a project manager or an expert. A sensor artefact that could have been removed at capture time is the hardest one to defend, because it was avoidable. The logic matches our guide to photogrammetry subcontracting for a chartered land surveyor : when a regulated professional signs off a deliverable, the acquisition chain must be traceable end to end.

Sideways flight along a vertical structure. Façade, dam, bridge pier, silo, water tower : the camera looks horizontally and the drone translates. This is where software correction is least comfortable and geometry most sensitive, while the resolution sought is at its finest. Our drone façade inspection page details the deliverables concerned.

Large format at sustained frame rate. On a 45 MP full-frame sensor, the data volume to read per image is considerable ; the Zenmuse P1 mechanical shutter, reaching 1/2000 s, is a design choice there, not a marketing option.

Conversely, four situations where requiring it means paying for nothing. Visual inspection first : when the deliverable is a photographic report rather than a metric model, frame geometry is not engaged — a crack is read, not measured in the site's coordinate frame. Stop-and-go next : if the drone halts at every capture point, ground speed is zero at trigger and rolling shutter produces exactly zero distortion, whatever the sensor. Slow surveys of small areas : at 3 m/s with a 30 ms readout sensor and a 2 cm GSD, travel during readout is 9 cm, about 4.5 pixels — above the threshold, but comfortably inside the correction model's validity domain on a steady nadir flight. Periodic construction monitoring finally, whose purpose is the relative comparison of successive states : what counts is protocol repeatability, and our drone construction monitoring page insists far more on registration between visits than on any component's performance.

Sensor size, lens and motion blur: the other half of the problem

The shutter monopolises the discussion, but it does not decide aerial image quality on its own. Two other parameters weigh at least as much : sensor size, and the lens projecting the image onto it.

Sensor size acts first through photosite area. At equal pixel count, a full-frame 35.9 × 24 mm sensor offers far wider photosites than a 1-inch one — 4.4 µm on the 45 MP Zenmuse P1. A wider photosite collects more photons in the same exposure time : signal-to-noise improves, and with it the local contrast fineness on which tie-point matching between frames depends. Irrelevant under bright June sun ; decisive on a November morning under overcast, on a north-facing façade, or late in the day as the flight window closes. The second effect is geometric : at equal focal length and height, a larger sensor covers a wider ground footprint, hence fewer flight lines and fewer frames for the same area. That is a direct productivity gain, partly offsetting the hardware premium.

The lens decides ground resolution and geometric fidelity. Focal length sets the ratio between flight height and pixel size projected on the ground — the whole subject of our guide to choosing ground sampling distance (GSD). Three trade-offs follow. A longer focal length holds a fine GSD while flying higher, hence further from obstacles and often in more comfortable airspace, at the cost of a narrower footprint and more flight lines. A short focal length covers ground fast but shows stronger edge distortion, which self-calibration has to estimate : that works well on an image block with proper overlap and good geometry, less well on a thin acquisition. And above all, a zoom is the enemy of a metric survey : lens calibration assumes stable interior parameters, whereas a variable focal length changes them at every reframing. Excellent for reading a serial plate or detailing a defect during inspection, to be banned from a photogrammetric block — a survey is flown at fixed focal length, or at a locked and documented one.

Motion blur, finally, is the most ordinary and most expensive trap. It has nothing to do with the shutter type : it comes from exposure time. The arithmetic fits in one line : at 10 m/s and 1/500 s, the drone travels 2 centimetres during exposure ; with a 2 cm GSD that is one pixel of smear, invisible. At 1/125 s the same flight travels 8 centimetres, four pixels : every fine detail in the image is spread out. And this is exactly what happens when light drops and automatic exposure lengthens the shutter to compensate — degradation arrives unannounced, just as the operator is thinking « let's finish the site before dark ». The study by T. Sieberth, R. Wackrow and J. H. Chandler, « Motion blur disturbs — the influence of motion-blurred images in photogrammetry », published in 2014 in The Photogrammetric Record (DOI 10.1111/phor.12082), documents that effect : blur caused by camera movement, arising as much from the drone's normal flight as from wind, turbulence or an abrupt pilot input, disturbs visual analysis of the data, introduces errors and degrades the accuracy of automatic photogrammetric processing. This is where a large sensor earns its place : it lets you keep a short exposure later into the day, preserving sharpness without pushing ISO to the point of drowning detail in noise. For flights in very low light, the procedure is set out in our guide to professional night drone flying.

The video case: the 180° rule, flicker and FPV artefacts

In video, the shutter is not a measurement problem but a look problem — no less binding when the deliverable is a corporate film or event coverage. Three points are enough to frame an order.

The 180° rule. Motion looks natural on screen when shutter speed is roughly double the frame rate : 1/50 s at 25 frames per second, 1/100 s at 50 fps. A faster shutter freezes each frame and produces a stuttering look, unpleasant on the aerial pans that are precisely the drone's added value. This is why serious aerial shoots carry neutral density filters : in daylight, without them, dropping to 1/50 s means overexposure. The absence of ND filters from a drone aerial video quotation says a lot about how the shoot was prepared.

Flicker under artificial light. The European 50 Hz mains makes luminaires without flicker-free drivers pulse at 100 Hz. A rolling shutter sensor reads its image while the light varies : the result is a series of dark and light horizontal bands rolling through the frame. The cure is to pick a shutter speed that is a whole multiple of the light period — 1/50, 1/100 or 1/200 s in Europe. That is compatible with the 180° rule at 25 fps, where 1/50 s satisfies both constraints. Less so at 24 fps, the cinema rate for which the rule gives 1/48 s, which does not sync with the mains : you then trade a little less motion blur against visible flicker. Worth checking before any shoot in an industrial hall, a sports hall, an underground car park or a showroom.

Rolling artefacts in FPV. In immersive flying, rotations are fast and deliberately expressive : this is where rolling shutter shows most, as leaning verticals during quick lateral passes and as the so-called « jello » effect, an image wobble that frame vibration amplifies. The cures are mechanical before they are digital : balanced propellers, gimbal dampers in good condition, clean gain settings. Software stabilisation in post corrects part of the residual wobble, never all of it. Our guide to FPV immersive flying and its professional uses covers the framework of these shoots, and our FPV immersive page the services concerned.

Specification, processing report and 2026 prices

Everything above boils down to a few lines to write into a tender, and a few lines to look for in the report you receive. Start with the tender. The wording to avoid is « drone fitted with a global shutter », which describes a technology almost no mapping drone carries and commits to nothing. Useful wording describes the expected result and leaves the contractor to choose the means :

• the accuracy expected on the deliverable, measured on independent check points not used in the computation, with a deviation table appended ;
• a commitment that residual rolling distortion stays negligible against that accuracy at the flight speed actually used — which forces that speed to be declared ;
• the target ground resolution and the fixed focal length used, with a ban on changing focal length within a single image block ;
• a minimum shutter speed compatible with ground speed and GSD, to bound motion blur, plus the accepted ISO ceiling ;
• forward and side overlaps, and what happens to frames captured in turns.

Now the processing report, which must be required and read. Five items are expected : the camera model and the shutter mode actually used during the mission (mechanical or electronic — the distinction is rarely noted spontaneously) ; a statement that the rolling shutter correction model was enabled, or why it was not ; the lens calibration parameters adopted, and whether they were self-calibrated or fixed ; the number of images discarded and why (blur, underexposure, failed matching) ; and finally the deviations on independent check points, kept separate from the control points used in the adjustment — a report showing only residuals on control points says nothing about real accuracy. These items are read alongside the deliverables described in our guide to photogrammetric survey deliverables ; and where the requirement concerns ground restitution under canopy, the sensor question becomes secondary to the technology one, covered in our LiDAR or photogrammetry comparison. On hardware choice itself, from the contractor's side, our guide to which professional drone to buy puts the shutter back among the other investment criteria.

2026 prices (excl. VAT), as a difference or a supplement on a survey job : a photogrammetric survey of an open 5 to 20 hectare area, mechanical-shutter drone with RTK georeferencing, orthophoto and digital surface model included, runs between €900 and €2,000. Requiring a large sensor with a mechanical shutter — a full-frame payload on a carrier aircraft where the contractor would otherwise have flown a compact multirotor — typically adds 20 to 40 % to the sortie, that is €300 to €900 : heavier hardware, shorter endurance per battery, more battery swaps, a more qualified operator. Conversely, keeping a rolling shutter sensor and lowering flight speed to stay under the distortion threshold stretches the mission by 30 to 60 % and costs €200 to €500 of extra flight time on a typical area — the hardware saving is thus partly paid back in the field. Processing with a rolling shutter correction model plus a deviation report on independent check points adds €150 to €400 where not already included. A full re-flight for unusable imagery — motion blur at the end of a dim day, uncorrected distortion caught at quality control — is commonly billed at 30 to 50 % of the original sortie, that is €300 to €800, not counting the downstream schedule slip. A travel charge applies beyond a 30 to 50 km radius. Request a quote stating the accuracy expected on the deliverable, the target ground resolution, the area and the nature of the structure : those four items, not the name of a shutter technology, determine the method and the price.

Frequently asked questions

Request a free quote

Put it into practice

• Drone aerial video: rates and cities covered from €500
• Aerial video in Boulogne-Billancourt Île-de-France
• Aerial video in Montreuil Île-de-France
• Aerial video in Nanterre Île-de-France

Also worth reading

• Olive orchards and drones: water stress and yield forecasting, price
• Prune d'Agen and drones: water stress and ripeness, price
• Asparagus and drones: spotting weeds on the ridge before emergence, price