A barcode that refuses to decode still sits somewhere in the image, and its position is often the only thing the downstream step needs. When the decoder gives up, a naive read call throws that geometry away with the string, and a pick, reject or rework station loses its target. This guide separates localization from decoding: which parts of a symbol survive a failed read, how many pixels each job needs, and how to get center, angle and bounding box out of a reader that returns nothing by default. It is written for line engineers integrating fixed code readers and smart cameras who must locate a barcode whose content is unreadable. Every number is checked by the arithmetic shown below.
Can you locate a barcode whose content will not decode?
Yes for QR and DataMatrix symbols: the finder patterns that frame the code carry its position independently of the payload, so a read that fails error correction can still return center, angle and bounding box; for 1D linear codes most readers return nothing.
Those are two different jobs sharing one image. Decoding runs error correction over the payload modules; localization needs only the symbol's frame. A decoder that fails the first can still succeed at the second, provided your integration lets the geometry through instead of discarding the whole result on a failed string. The QR code finder pattern detection: locate before decoding guide walks that localization half step by step.
Which geometry survives when the payload fails?
The finder structure and the quiet-zone border survive damage that destroys the payload: a QR code keeps three 7 × 7 module finder patterns in its corners and a DataMatrix keeps its solid L-shaped border and timing tracks, none of which error correction protects or needs.
On a QR code those finders are three nested squares, one in every corner except the bottom right, each built from a 1:1:3:1:1 sequence of dark and light rings. The geometry is redundant by design: two finder centers fix rotation, the third exposes skew, and together the three fix the center. The outer corners give the bounding box.
A DataMatrix carries its frame as a solid L along the left and bottom edges plus an alternating timing track along the top and right. Two straight, high-contrast edges fix orientation and extent, which is why symbology documentation describes the finder pattern as defining the symbol's shape, size and X-dimension.
Error correction never touches this frame; it protects payload codewords only. That division of labor is the whole trick: print damage can eat a large share of the data and still leave every corner intact. The quiet zone is the backup — even with one finder scuffed, a clean border still bounds the symbol.
How many pixels does localization need compared with decoding?
About 3 pixels per module locates the finder patterns, while 5 pixels per module is the comfortable floor for decoding the data region, so a symbol can sit well inside localization range and still fail to read.
Pixel count, not contrast, decides which of the two jobs a sensor can do. The 3 px floor comes from the finder's 1:1:3:1:1 run-length ratio, which smears into noise below that; decoding wants more because every narrow element, not just the frame, must resolve.
Worked example: a field of view 120 mm wide, one QR version 2 symbol printed 20 mm on a side. Version 2 carries 25 × 25 modules, so one module measures 20 / 25 = 0.8 mm. Localization at 3 px per module needs a sensor width of 120 × 3 / 0.8 = 450 px. Decoding at 5 px per module needs 120 × 5 / 0.8 = 750 px. A 2560 px sensor delivers 0.8 × 2560 / 120 = 17.1 px per module here, past both thresholds with room for focus and print variation.
The same numbers set accuracy. Version 2 finder centers sit 18 modules apart, so 18 × 0.8 = 14.4 mm. A half-pixel center error is 0.5 / 17.1 = 0.029 mm, which gives an angle error of atan(0.029 / 14.4) ≈ 0.12°. A pick or reject guided by that pose is limited by the mechanics, not by the failed decode.
Size your own window from field of view and module size:
The LS-CRD-0028 is that sensor: 2560 × 2048 px, 30 fps and 60 codes/s, with an 8 mm lens on a 5 MP global-shutter reader.
- Interface
- Fast Ethernet
- Resolution
- 5 MP
- Sensor format
- 1/1.7"
- Shutter
- Global
Why does a reader fail the decode but still see the symbol?
Decoding fails when damaged or low-contrast modules exceed the error-correction budget, while the finder patterns keep their shape, so the symbol remains visible geometry the decoder cannot convert to a string.
Keep the two failure families apart:
- Print damage: a blocked inkjet nozzle or a dead thermal element wipes one region of the symbol, and ink spread closes the narrow bars. QR error correction rebuilds roughly 7% of codewords at level L and about 30% at level H; past the budget the string fails while the corners still read.
- Optical damage: specular glare, uneven lighting or motion blur removes contrast across the whole label, and detection degrades along with the decode.
Glare is the fixable one. Cross-polarize: a polarizing film on the light at 90° to a film on the lens removes the specular reflection and keeps the printed, diffuse contrast. Crossed polarizers for specular glare suppression in machine vision shows the set-up, and the polarized MV-IDB005EXI-05NRP applies the same filter pair with its own light, focused at 70 mm.
Give the symbol room as well: the quiet-zone border is part of its address, and the QR code quiet zone and integer module scaling guide has the margin math. On a moving belt, set exposure from belt speed and px per module; Barcode capture on a moving vehicle: speed, blur and read rate works through that trade. If the lamp must move or the film sits off-axis, lighting accessories covers extension cables and diffuser plates.
How do you get a reader to return position with empty content?
Look for a return-errors switch in your decoding library: one widely used option is a returnErrors flag in ReaderOptions, aliased DecodeHints, which returns a result with empty content, error information set and the position filled in for 2D matrix codes.
Integration code usually tests two states, got a string or got nothing, and that is where the geometry dies. Branch on three instead: decoded, with content; located but unread, with content empty and position and error fields set; not found, with no result at all. Log the middle state as an event, not as silence.
Commercial fixed code readers expose the same idea through their own result fields: a detection, score or corner list that stays valid when the string field comes back empty. Check the result structure before assuming a failed string means a failed detection. The barcode detection pipeline fallbacks article explains what those pipelines attempt in order, and barcode read confidence shows how much weight a marginal result carries.
For 1D linear codes this switch is a 2D feature: linear symbologies have no independent finder structure for the library to report, so a failed scan generally returns nothing at all.
What can you do with a position and no string?
Plenty: steer a reject or rework station to the symbol's center, guide a pick by the label's known offset from the part, or open a tight window around the bounding box for a second, higher-resolution read attempt.
Guidance: when the label sits at a fixed offset from a part feature, the symbol's pose maps straight to part pose. The mapping lives in your hand-eye calibration, and the barcode position accuracy for pick and place guide sizes the chain from pixels to millimetres. One acquisition can carry both the part location and the code attempt; part position and barcode reading in one image lays out that pipeline.
Routing: a reject pusher needs only the center coordinates, and the angle tells a flipper or a picker how the part lies on the belt.
Second attempt: draw a tight region around the bounding box and retry, because a staged reader at higher px per module recovers codes the wide window cannot. The ROI for inline code reading guide sizes that window.
When is position-only localization the wrong answer?
Not acceptable as a pass: a code that locates but cannot decode is still a failed read on any traceability or compliance line, and the position's job there is to route the part out, not to stand in for the string.
Verification treats the matter the other way up: print-quality standards score an undecodable symbol as a failure regardless of how cleanly the frame was found, so compliance still needs the payload.
Localization has its own failure mode. QR pose leans on all three finders, so a scuffed corner degrades center and angle before it disappears entirely; treat a located-but-unread result with the same suspicion as a marginal string, and check that the corners agree on one consistent frame before trusting the pose.
Keep the evidence: a full-resolution snapshot around the located box lets an operator re-key the string or lets an offline decoder retry off-shift. The barcode inspection image archiving to cloud storage guide covers where those images go.
Which reader should you pick for geometry on failed reads?
Choose a reader whose result exposes center, angle and bounding-box fields independently of the payload string, with enough pixels per module for your smallest code: 1.3 MP units cover tight fields, while 12-20 MP units hold localization across wide logistics windows.
These three fixed code readers span that range.
| Specification | MV-ID3013PM-12M-WBN | MV-ID5120RM-08M-RBN | MV-ID6200PM-00C-NNG |
|---|---|---|---|
| Symbologies | 1D codes: Code 39, Code 93, Code 128, CodaBar, EAN 8, EAN 13, UPCA, UPCE, ITF 14, ITF 25, Matrix 25, MSI, China Post, Code 11, and Industrial 25; 2D matrix codes: QR Code (GS1-QR included), Data Matrix (GS1-DM included), MicroQR, AZTEC, HanXin, Maxi code; Stacked codes: PDF417, MicroPDF417 | 1D codes: Code 39, Code 93, Code 128 (GS1-128 included), CodaBar, EAN 8, EAN 13, UPCA, UPCE, ITF 14, ITF 25, Matrix 25, MSI, China Post, Code 11, Industrial 2of5, and Pharmacode; 2D codes: QR Code (GS1-QR included), Data Matrix (GS1-DM included), MicroQR, AZTEC, HanXin; Stacked codes: PDF417, MicroPDF417 | 1D code: Code 39, Code 128, Code 93, ITF25, CodaBar, EAN8, EAN13, UPCA, UPCE, ITF14, etc.; Matrix 2D code: QR Code, Data Matrix, etc.; Stacked 2D code: PDF417 |
| Max. frame rate | 60 fps | 10 fps | 20 fps |
| Max. reading speed | 84 codes/sec | 30 codes/sec | 60 codes/s; 30 codes/s |
| Sensor type | Global shutter, mono, 1/2.7", 4 μm × 4 μm | Rolling shutter, mono, 1/1.57", 2 μm × 2 μm | CMOS, global shutter |
| Resolution | 1280 × 1024 | 4096 × 3024 | 5120 × 3968 |
| Exposure time | 35 μs to 1 sec | 33 μs to 1 sec | 16 μs ~ 0.5 sec |
| Gain | 0 dB to 28 dB | 0 dB to 34 dB | Analog gain: up to 5×; Digital gain: -6 to 6 dB; Analog gain: up to 5×; Digital gain: -24 to 24 dB |
| Data interface | Fast Ethernet (100 Mbit/s) | Gigabit Ethernet (1000 Mbit/s) | Gigabit Ethernet (1000 Mbit/s) |
| Communication protocol | SmartSDK, TCP Client, TCP Server, Serial, FTP, Profinet, Ethernet/IP, MELSEC/SLMP, ModBus, UDP, Fins, Modbus RTU | SmartSDK, TCP Client, Serial, FTP, TCP Server, Profinet, MELSEC/SLMP, Ethernet/IP, ModBus, ModBus RTU, UDP, Fins | SmartSDK, TCP Client, Serial, FTP, TCP Server, UDP |
| Digital I/O | 17-pin M12 connector provides power and I/O, including non-isolated input (LineIn 0/1/2) × 3, non-isolated output (LineOut 3/4/5) × 3, RS-232 × 1. Supports device triggering via pressing button on top.;17-pin M12 connector provides power and I/O, including isolated input (LineIn 0/1/2) × 3, isolated output (LineOut 3/4/5) × 3, RS-232 × 1. Supports device triggering via pressing button on top. | 12-pin M12 connector provides power and I/O, including opto-isolated input (LineIn 0/1/2) × 3, opto-isolated output (LineOut 3/4/5) × 3, RS-232 input × 1, and RS-232 output × 1.; Supports device triggering via pressing button on device and smart tune. | 12-pin M12 connector provides power and I/O, including opto-isolated input (LineIn 0/1/2) × 3, opto-isolated output (LineOut 3/4/5) × 3, and RS-232 × 1. |
| Power supply | 24 VDC | 24 VDC | 12 ~ 24 VDC |
| Power consumption | Avg.: 9 W @ 24 VDC (light source enabled); Max.: 48 W @ 24 VDC (light source enabled) | Avg.: 15.5 W @ 24 VDC (light source enabled); Max.: 36 W @ 24 VDC (light source enabled) | – |
| Price | $1,979.00 | $1,969.00 | $5,699.00 · quoted |
All 3 share: Humidity 20% RH to 95% RH (no condensation); Client software IDMVS. Values as printed in each product's data sheet; open a part for its drawing and full specifications.
Compact 1.3 MP readers put a dense pixel grid on a small, tightly presented code. High-resolution readers keep px per module up across a wide window, where the code can sit anywhere on a box face. If the same station must also verify presence or measure the part, smart cameras run locate-and-inspect logic on board without a PC.
To size your own window, walk the worst label through the arithmetic above: module size from the symbology version, field of view from part wander, then required sensor width. The code reader finder filters readers by resolution, shutter and interface, the system configurator pairs the chosen reader with lens, light and cabling, and the applications team answers integration questions through the contact page. The code-reading topic page collects the rest of the decoding guides.
Questions
How much quiet zone does a barcode need for reliable position detection?
Keep 4 modules of quiet zone around a QR code and at least 1 module around a DataMatrix. Localization keys off the clean border and the finder corners, so artwork, a fold edge or a neighboring label inside that margin destroys exactly the geometry a position-only result depends on.
Does motion blur kill localization before it kills the decode?
Usually the opposite. Blur smears module edges, and the decode needs every narrow element resolved, while each finder pattern spans 7 modules and tolerates far more smear. At 300 mm/s with a 2 ms exposure the image smears 300 × 0.002 = 0.6 mm, about 0.75 of a 0.8 mm module: enough to break most decodes, not the corners.
What should I record when a read returns position but no string?
Log the timestamp, center coordinates, angle, bounding box, the reader's error or confidence field and the full-resolution image, then treat the event as a failed read for traceability. The saved image lets an operator key the string in by hand or lets an offline decoder try again at full quality.
Is a located-but-unreadable code acceptable on a traceability line?
No. Print-quality grading such as ISO/IEC 15415 scores an undecodable 2D symbol as a failure regardless of position, so a located geometry must route the part to reject or rework rather than stand in for the decoded string. Localization steers and diagnoses; traceability still needs the payload.
Can I locate a 1D linear barcode the same way when it fails to read?
Generally not through the decoder. Linear symbologies have no finder structure that reader libraries report on failure, so the return-errors switch is a 2D feature. You can still find the pattern yourself: a strong run of alternating bar edges in the image gives a center and tilt, but that is your own geometry code, not a decode result.
