Can a phone running Dragon Vision® see the same vibration as a dedicated motion amplification camera? The physics says yes below about 300 Hz, and above that nobody sees much
This is the long version: number by number, with every source and every estimate marked. If you want the plain-language answer first, read Do you need a motion magnification camera, or is your phone enough?
Short answer. Video-based vibration analysis measures displacement, and displacement is limited by two things a spec sheet rarely states plainly: how much of the machine fits in the frame and how much light reaches each frame. Distance by itself does not matter, and neither does the frame rate beyond a point. Once you compare on those terms, Dragon Vision® on an iPhone 15 Pro Max with its 5× telephoto lens resolves displacements in the same order of magnitude as a dedicated camera built on a 1.9 MP industrial sensor, and it samples faster in full HD (240 fps against 180 fps). A dedicated high-speed camera does reach higher frequencies, but only by shrinking the picture to a strip or a patch and starving it of light, in a band where a real machine produces almost no displacement at all. Below roughly 300 Hz, which is where unbalance, misalignment, looseness, resonance and structural problems live, the phone in your pocket is the right instrument. Above it, the right instrument is an accelerometer, not a faster camera. Every estimate in this article is marked as such, and the method is at the end.
We make one of the two products compared here, so the numbers for our own side are the ones we published and measured, and the numbers for the other side come from its public manuals, spec sheets and the camera vendor’s own data sheets, as of September 2026. Where a figure had to be derived rather than read, it says estimate, and the section How these numbers were obtained shows the arithmetic so you can redo it with your own camera. Motion Amplification® is a registered trademark of RDI Technologies; in this article the technique in general is called motion magnification, the name it was given by the researchers who invented it.
Phone vs dedicated camera at a glance
✔ yes · — no · ≈ estimate (see method)
| Dragon Vision® + iPhone 15 Pro Max | Dragon Vision® + Galaxy S10 | Dedicated 1.9 MP camera (Sony IMX174) | Dedicated high-speed camera (Fastec HS5i) | |
|---|---|---|---|---|
| The camera | ||||
| Sensor | 48 MP main + 12 MP 5× telephoto (120 mm eq.) | 12 MP main (26 mm eq.) | Sony IMX174, 1920×1200 mono, global shutter, 5.86 µm pixels | Fastec HS5i, 2560×2048, 12-bit, 5 µm pixels |
| Video at full HD | 1080p at 240 fps, unlimited length | 1080p at 240 fps | 1080p at 120 fps default, up to 180 fps | 1080p at up to 1,400 fps |
| Fastest mode | 240 fps at 1080p (no faster mode) | 960 fps at 720p, 0.4 s bursts | 1,300 fps, ≈150 rows high | 29,000 fps at 64×32 pixels |
| What the frame rate buys | ||||
| Max frequency in full HD (Fmax) | 120 Hz (7,200 CPM) | 120 Hz (7,200 CPM) | 60 Hz default, 90 Hz max (5,400 CPM) | 700 Hz (42,000 CPM) |
| Max frequency in the fastest mode | 120 Hz | 480 Hz, in a 720p frame | 650 Hz, in a strip ≈22 cm × 1.8 cm at 1 m | 14,500 Hz, in a patch ≈6 mm × 3 mm at 1 m |
| Light per frame in the fastest mode vs 120 fps | ÷2 | ÷8 | ≈ ÷11 | ≈ ÷240 |
| Displacement resolution at 1 m | ||||
| Width of the frame at 1 m | ≈1.5 m at 1×, ≈0.30 m at 5× | ≈1.4 m (1080p or 720p) | ≈0.23 m with the 50 mm lens | ≈0.19 m with the 50 mm lens |
| Micrometres per pixel at 1 m | ≈ 780 at 1×, 156 at 5× | ≈ 720 at 1080p, 1,080 at 720p | 117 with the 50 mm lens | 100 with the 50 mm lens |
| Smallest displacement, published | 2.5 µm (0.1 mil) at 1× on an iPhone XS, measured | 2.5 µm (0.1 mil) at 1080p, same measurement | 2.5 µm (0.1 mil) in the 2018 manual; 0.25 µm (0.01 mil) on the 2026 product page, same camera and lens | 0.25 µm (0.01 mil), 0.125 µm at close focus |
| Smallest displacement at 1 m, with the best lens | ≈ 0.55 µm (0.02 mil) at 5× | ≈ 2.5 µm (0.1 mil) at 1080p; 3.8 µm (0.15 mil) at 720p | 0.25 µm (0.01 mil), as published | 0.25 µm (0.01 mil), as published |
| Sub-pixel factor (frame width ÷ pixels ÷ resolution) | ≈280, measured | ≈280, measured | ≈47 in the 2018 manual; ≈470 on the 2026 page | ≈ 400 |
| Cost and workflow | ||||
| Software price, published | $4,499.95 one-time, no annual fee | $4,499.95 one-time, no annual fee | Quote only | Quote only |
| Hardware you need to buy | None if you own the phone; the phone otherwise | None if you own the phone | Camera, lenses, acquisition laptop (i7, 32 GB RAM, 1 TB SSD), LED light kit | Same, plus the high-speed camera |
| Records handheld | ✔ | ✔ | — tripod | — tripod |
| Processes video from any camera | ✔ iOS, Android, Windows, online | ✔ | — its own camera | — its own camera |
| Exports spectra (UFF 58, ASCII) | ✔ | ✔ | ✔ CSV | ✔ CSV |
| Continuous monitoring option | — | — | ✔ fixed-camera model | ✔ fixed-camera model |
| Capabilities, out of the yes/no rows | 3 / 4 | 3 / 4 | 2 / 4 | 2 / 4 |
Dragon Vision® resolution figures are scaled from a published measurement on an iPhone XS (0.1 mil at 1 m in full HD) using each phone's field of view. Dedicated-camera figures come from RDI's Iris M™ user manual (2018), its Iris MX™ product sheet (2024) and the data sheets of the cameras inside them (FLIR Grasshopper3 with a Sony IMX174 sensor; Fastec HS5i). Window sizes and light per frame are derived from those sheets. Prices as of September 2026.
Read the table from the middle band down. The dedicated cameras win the fastest-mode row by a mile, and the next two rows explain why that row is worth much less than it looks: the fast modes come with a picture the size of a strip or a stamp, and with a fraction of the light. The resolution band shows the number that actually decides what you can see, and there the four columns are within a factor of two to three of each other once the phone uses its telephoto lens.
2. The frequency ceiling you cannot fill
Frame rate sets the highest frequency a video can resolve: half the frames per second. That is the number every spec sheet leads with, and it is the least useful number in the comparison, because a camera measures displacement, and displacement collapses with the square of the frequency.
For a sinusoidal vibration of acceleration a at frequency f, the peak displacement is x = a ÷ (2πf)². Vibration analysts think in g at high frequency because that is how bearings and gears are measured, so here is what one g of acceleration looks like in displacement, against the best displacement floor any of the four systems publishes:
| Frequency | Displacement of 1 g | Dedicated camera floor, 0.25 µm | iPhone 5× floor, ≈0.55 µm |
|---|---|---|---|
| 10 Hz | 2,485 µm | visible, 10,000× the floor | visible |
| 60 Hz | 69 µm | visible | visible |
| 120 Hz | 17 µm | visible | visible |
| 300 Hz | 2.8 µm | 11× the floor | 5× the floor |
| 480 Hz | 1.1 µm | 4× the floor | 2× the floor |
| 650 Hz | 0.59 µm | 2× the floor | below the floor |
| 1,000 Hz | 0.25 µm | at the floor | below |
| 2,000 Hz | 0.06 µm | 4× below the floor | below |
| 5,000 Hz | 0.01 µm | 25× below the floor | below |
One g of vibration, drawn against the two displacement floors on logarithmic axes. With a tenfold margin the line crosses the dedicated camera’s floor at about 300 Hz and the phone’s at about 200 Hz. Beyond that point the difference between the two cameras is academic: neither sees the machine.
Turn the table around and ask how many g a machine would need before the camera could even touch its floor at high frequency: 1 g at 1,000 Hz, 4 g at 2,000 Hz and 25 g at 5,000 Hz. A usable measurement wants a signal ten times its floor, so multiply by ten. A machine vibrating at 250 g at 5 kHz is not being monitored; it is being demolished.
That is the whole high-frequency argument in one line: the useful ceiling of video-based vibration is set by the displacement floor, not by the frame rate, and with one g of vibration and a tenfold margin it sits around 300 Hz for the dedicated camera and around 200 Hz for the phone. The difference between 120 Hz and 700 Hz of nominal Fmax is real; the difference in what you will actually see on a machine is small. Bearing defects, gear mesh and everything else that lives in the kilohertz belong to an accelerometer, which is why our own FAQ has always said so, and why a wireless sensor on the bearing housing costs a fraction of a high-speed camera.
4. What the phone gives up, honestly
A fair article lists its own side’s limits in the same tone.
- Compression and colour. The phone records compressed HEVC video from a colour sensor with a rolling shutter; the dedicated camera records uncompressed monochrome frames with a global shutter. Those are real advantages for sub-pixel tracking, and they are the most likely reason the dedicated camera’s current published factor (about 470) is above our measured one (about 280). Our measurement was made on a 2018 phone; newer sensors, 10-bit capture and the lower distortion of a telephoto lens all push the other way, but until we repeat the measurement on a current phone the 0.02 mil figure stays an estimate.
- Optical stabilisation. A phone’s telephoto module physically moves its lens to cancel hand shake, which is motion the machine did not make. Dragon Vision® already filters the low-frequency movement of a hand, and a gimbal removes most of the rest; whether the stabiliser needs to be disabled on a tripod is something we will report with the measurement above.
- 960 fps on a phone is not a high-speed camera either. The Galaxy S10 records real 960 fps, but at 720p, in bursts of 0.4 s (an 0.8 s mode exists that records 480 fps and interpolates to 960; interpolated frames invent motion and must never be used for vibration). At 720p the pixel is 1.5 times larger, so the floor rises to about 3.8 µm (estimate), and 0.4 s of video gives a spectrum with 2.5 Hz lines and nothing to average. Its 480 Hz ceiling has the same problem as the dedicated camera’s: at 480 Hz, one g is 1.1 µm, below that floor. Our advice with any phone is the same as our advice about the dedicated high-speed camera: use 240 fps at 1080p, where you can record for minutes and the frame is the whole machine. The extra fps do not buy signal.
- No continuous monitoring. A phone is an instrument you walk with. If you need a camera watching a machine permanently, the dedicated-system vendor sells one and we do not; our permanent instruments are wireless sensors.
- Which phone. The estimates for the 5× lens apply to the iPhone 15 Pro Max and to the 16 Pro and later, which carry a 120 mm-equivalent telephoto that records 240 fps. The 15 Pro has a 77 mm 3× lens (about 0.03 mil, estimate), and the non-Pro models have no telephoto, so they stay near the 0.1 mil we measured. Any Android phone works as long as it records 240 fps at 1080p; Dragon Vision® runs on iOS, Android and Windows and the online version processes video from any of them.
What it costs
Dragon Vision® for Windows has a published price, $4,499.95 one-time with no annual fee, and the Windows licence includes the mobile app; the camera is the phone you carry, or a used one bought for the purpose. A dedicated motion amplification system is sold on quotation, and the quotation is for a system: the camera, a set of lenses (6, 12, 25, 50 and 100 mm), a ruggedised acquisition laptop with 32 GB of RAM and a terabyte of storage, an LED light kit, a tripod, training, and on the current platform a modular software licence where route collection is standard and rotation, high-speed, accelerometer, modal and limit modules are upgrades. There is also an active rental market for these cameras, which tells you something about where the purchase price sits. Users on a public analysts’ forum have reported paying roughly 70% less for Dragon Vision® than for the dedicated system; we cannot verify their quotations, so treat that as a user report rather than a price list.
The right comparison is not the two price tags. It is what each one is for: a phone-based system is bought so that every analyst on a site can carry motion magnification in a pocket and use it on the first suspicious machine of the day; a dedicated camera is bought for a specialist, a case, and a tripod. Both are legitimate. Only one of them costs less than the accelerometers you should be buying for the high frequencies anyway.
FAQs about phone-based motion magnification
Is Dragon Vision® the same thing as Motion Amplification®?
They are two commercial products built on the same principle: track the position of edges in a video to a fraction of a pixel, turn that into a displacement signal for every point in the frame, and optionally exaggerate the motion so the eye can see it. Motion Amplification® is the registered trademark of RDI Technologies for its system, which pairs the software with its own camera. Dragon Vision® is ERBESSD INSTRUMENTS®’ software, which processes video from any camera, including the phone you already own, and exports the resulting spectra to Digivibe MX®, ASCII and UFF 58. The technique in general is called motion magnification.
Can a phone really be as sensitive as a dedicated motion amplification camera?
Within a factor of two, yes, on the published numbers, provided you frame the same area. The resolution of any video method is the size of a pixel on the machine divided by the sub-pixel tracking factor. The dedicated camera’s 50 mm lens frames about 23 cm at 1 m; the 5× telephoto of an iPhone 15 Pro Max frames about 30 cm at the same distance. With the tracking factor we measured on an iPhone XS, that gives an estimated 0.5 µm (0.02 mil) against the dedicated camera’s published 0.25 µm (0.01 mil). The same dedicated camera was specified at 2.5 µm (0.1 mil) with the same lens in its 2018 manual, which the phone measurement already beats.
Why does a camera struggle with high-frequency vibration?
Because a camera measures displacement, and for a given acceleration the displacement falls with the square of the frequency. One g of vibration is 69 µm of movement at 60 Hz, 0.25 µm at 1,000 Hz and 0.01 µm at 5,000 Hz. The best published displacement floor of any motion amplification camera is 0.25 µm, so at 1,000 Hz a machine needs a full g just to reach the floor and about 10 g for a clean measurement, and at 5,000 Hz it needs hundreds of g. Real machines do not do that. Above a few hundred hertz the instrument that sees the fault is an accelerometer.
Does a phone that records 960 fps replace a high-speed camera?
No, and neither does the high-speed camera replace an accelerometer. A Galaxy S10 records real 960 fps only at 720p and in 0.4 s bursts, which raises the displacement floor and leaves a coarse spectrum with nothing to average. Its 480 Hz ceiling lands where one g is about 1 µm, below that floor. Record at 240 fps in 1080p instead: you get 120 Hz of bandwidth, the whole machine in the frame, and minutes of video. Avoid any “super slow motion” mode that interpolates frames; invented frames invent frequencies.
Does video compression ruin the measurement?
It costs some precision, not the measurement. Phones record compressed HEVC video and a dedicated camera records uncompressed frames, which is one reason the dedicated camera can claim a higher tracking factor. In practice, our 0.1 mil measurement at 1 m in full HD was made on compressed phone video, and it resolves the displacements that matter below 300 Hz with margin. Use the highest bitrate your phone offers, avoid digital zoom that is not backed by a real lens, and keep the machine large in the frame.
Can I use an Android phone with Dragon Vision®?
Yes. Dragon Vision® runs on iOS, Android and Windows, and the online version processes video recorded on any of them. What matters is the recording: 1080p at 240 fps, a real optical lens rather than digital zoom, and a machine that fills the frame. On the Windows and online versions you can also process video from action cameras and camcorders, with the caveat that fisheye lenses reduce amplitude accuracy.
How do I check that a phone measurement is right?
Put an accelerometer on the point you are filming and compare. Dragon Vision® was calibrated against a NIST-traceable accelerometer, and the same check takes five minutes on site: mount a wireless sensor on the bearing housing, film the housing at 240 fps for twenty seconds, and compare the displacement spectrum from the video with the one from the sensor at the running speed and its first harmonics. Agreement at those frequencies is what you should expect; disagreement usually means the frame was too wide or the light too low.
Should I buy a dedicated camera, rent one, or use software on a phone?
Buy the phone software if you want every analyst on site to carry motion magnification and use it on the first suspicious machine of the day; at a published one-time price with no annual fee it costs less than a set of accelerometers. Rent a dedicated camera for a one-off study that needs a fixed, tripod-mounted, uncompressed recording, or when a customer’s procedure names that system. Buy the dedicated camera when you need continuous video monitoring of a specific machine, which is the one capability a phone cannot offer.
How these numbers were obtained
So that anyone can check or redo them.
- Frame widths are focal length arithmetic: width at 1 m = 36 mm ÷ 35 mm-equivalent focal length, for the phones (26 mm on the Galaxy S10 and the iPhone XS, 24 mm and 120 mm on the iPhone 15 Pro Max); and sensor width × 1,000 ÷ 50 mm for the dedicated cameras (11.25 mm for the Sony IMX174; 12.8 mm for the full Fastec HS5i sensor, of which the 1080p window uses 1920 of the 2560 columns, hence 19 cm).
- Micrometres per pixel = frame width ÷ horizontal pixels (1920 for 1080p, 1280 for 720p, 1920 and 2560 for the dedicated cameras).
- The sub-pixel factor of Dragon Vision®, about 280, is our published 0.1 mil (2.54 µm) at 1 m in full HD on an iPhone XS divided into that phone’s 720 µm pixel. Every Dragon Vision® estimate in this article applies that same factor to a different frame width; it does not assume any improvement from newer phones.
- The sub-pixel factors of the dedicated camera are its published resolutions divided into its 117 µm pixel at 1 m with the 50 mm lens: 2.5 µm in the 2018 Iris M™ user manual, 0.25 µm on the 2026 product page.
- Window sizes at high frame rates come from the Fastec IL5/HS5i data sheet, which lists frames per second for each picture size; the 150-row estimate for the Iris M™ at 1,300 fps assumes the same pixel throughput as its 1920×1200 frame at 163 fps.
- Light per frame is one over the frame rate; the floor degradation assumes shot-noise-limited tracking, so it scales with the square root of the light ratio. This is the roughest estimate in the article and is marked as such.
- Displacement of 1 g is 9.81 ÷ (2πf)² at each frequency.
What is still pending is the measurement that turns the 0.02 mil estimate into a number: the same test we ran on the iPhone XS, repeated on an iPhone 15 Pro Max at 240 fps with the 1× and 5× lenses against a NIST-traceable accelerometer, with the real frame width recorded. This article will be updated with the result.
