Final year capstone build, 1,600 test photos

Capstone project poster example

A capstone project poster shows a thing that was built and tested against a target set at the start. This A0 portrait example comes from a final year engineering team of four and their supervisor: a $38 clip on camera that reads old water meters, tested on 1,600 photos of three meter models, reading 1,539 correctly, 96.2% against a 95% target.

Create a poster with OneCraftA0 Portrait, printed at 841 by 1189 millimetres

The whole board

The poster at full size, exactly as it prints. Every number, citation and caption on it was written for this example, so the layout is being judged on real content.

PIT
Reading old water meters with a $38 camera: a final year capstone build tested on 1,600 meter photos
L. Haverford ¹, M. Oduya ¹, J. Petrakis ¹, R. Sundqvist ¹, Dr K. Ellwood (supervisor) ²
1 Final year capstone team, Bachelor of Engineering, Pellmarsh Institute of Technology · 2 School of Electrical and Computer Engineering, Pellmarsh Institute of Technology
The brief
Many homes still have mechanical water meters with a row of rolling digits, and someone has to walk to each one and write the number down. Smart meters fix this but cost hundreds of dollars a home to fit. Our brief from a regional water utility was narrower: build a clip on device under $40 in parts that photographs the existing meter, reads the digits on the device itself and sends one number a day, without replacing the meter or the plumbing. Success meant reading at least 95% of meters correctly to every digit, across the meter models the utility has in the ground, in daylight and in the dark of a meter pit.
WHAT WE DID
Device. A low cost camera board with an infrared light, a 3D printed clip and a battery pack: $38 in parts.
Photos. 1,600 photos of 3 meter models the utility uses, A, B and C, in daylight and in low light.
Labels. Every photo hand labelled by two team members; the 23 disagreements settled by the supervisor.
Reader. Finds the digit window, splits it into 6 digits and classifies each with a small network on the device.
Test. A reading counts as correct only if all 6 digits match the label. Time measured on the device itself.
What we found
The reader got 1,539 of 1,600 readings right to every digit, 96.2%, which clears the 95% bar. Daylight beat low light on every model: model A 297 of 300 against 248 of 260, model B 275 of 280 against 222 of 240, model C 274 of 280 against 223 of 240. The 61 misreads had four causes: glare on the glass 24, a digit caught half way through rolling 19, dirt on the glass 12 and other faults 6. Each reading took 0.4 s on the device, so a daily reading costs almost no battery.
Daylight
Low light
80%85%90%95%100%
Figure 1. Readings correct to every digit, %, by lighting. Bars left to right: meter models A, B and C. 1,600 photos.
Source: team test set, 1,600 labelled photos.
Next semester
Three jobs carry over. First, glare: 24 of the 61 misreads came from light bouncing off the meter glass, so we will test a polarising film over the lens and move the infrared light off axis. Second, rolling digits: 19 misreads caught a digit half way between two numbers, and a second photo taken a few seconds later should settle most of them, at the cost of a little more battery. Third, a field pilot: 20 devices clipped to real meters for four weeks, with the utility reader still doing the rounds so every device reading can be checked against a person. If the field accuracy holds near 96%, the utility will consider a larger trial.
REFERENCES
1.Iwasaki, T. and Morrow, D. (2023) Digit recognition on rolling counter meters. Journal of Embedded Vision, 7(2), 58 to 71.
2.Brennan, C. (2024) Low light imaging in meter pits. Utility Sensing Letters, 3(1), 14 to 22.
3.Ellwood, K. (2025) Capstone design reviews: setting a test bar before the build. Pellmarsh Engineering Education Note 1, 2025.
1,539 of 1,600
readings correct to every digit
96.2%, against a target of 95%; 61 misreads across 3 meter models
06121824
Glare
Rolling digit
Dirty glass
Other
Figure 2. The 61 misreads by cause: glare 24, a rolling digit 19, dirty glass 12, other 6.
Source: team review of every misread photo.
Result
A $38 clip on reader met the utility target of 95%, reading 96.2% of meters correctly.
1,600 meter photos, 3 models, hand labelled
96.2% of readings correct to every digit
0.4 s per reading on the device
$38 in parts per device

Block by block

What each block on the board is for, in the order a reader walks it.

Title band and five names
The title names the device, its cost and the test. Four students and the supervisor, marked as such, over two affiliations at the fictional Pellmarsh Institute of Technology.
Brief, method and findings columns
Three columns with icon markers: the utility brief and the 95% target, five method entries on device, photos, labels, reader and test, then results by meter model and lighting.
Figure 1 and key number
Bars give the share read correctly in daylight and low light for meter models A, B and C. The stat block gives 1,539 of 1,600.
Figure 2
Horizontal bars break the 61 misreads into glare 24, rolling digit 19, dirty glass 12 and other 6.
Next semester
The three next steps from the row are written as a section: a polarising film for glare, a second photo for rolling digits and a 20 device field pilot.
Result callout, key points and references
The callout states the target was met. The four key points, 1,600 photos, 96.2%, 0.4 s and $38, sit in the checked list, and three references fill the left foot.

What makes this board work

A target set before the build

The brief states the 95% bar the team set in advance. 96.2% then reads as a pass against the client brief, not a number chosen after the fact, which is what capstone assessors look for.

Every misread is accounted for

1,600 less 1,539 is 61, and Figure 2 splits all 61 by cause: 24 plus 19 plus 12 plus 6. Next semester then works through the two largest causes in order.

Results split by condition

Daylight beat low light on all three models, 99.0% against 95.4% on model A. The split shows where the reader fails, a meter pit at night, which a single 96.2% would hide.

Questions people ask

How is a capstone poster different from an undergraduate research poster?

A capstone board reports a build tested against a client or course brief, so it leads with the problem, the design and whether the target was met. A research poster reports a study and leads with a question. This board opens with the utility brief and the 95% bar, and closes with next semester.

How do I credit a supervisor on a capstone poster?

List them last and mark the role, as this board does with Dr K. Ellwood (supervisor), with their school as a separate affiliation. The four students share the capstone team affiliation. That tells an assessor at a glance whose work the poster shows and who guided it.

Where do the four key numbers go on this layout?

In the four point checked list near the foot: 1,600 photos, 96.2%, 0.4 s and $38. The three round icons on this design are heading markers beside the first three columns, not number blocks, so they cannot carry the key points.

Should a capstone poster show what went wrong?

Yes, and with numbers. Assessors and industry visitors trust a board that names its failures. Here Figure 2 counts the 61 misreads by cause and the next semester section plans a fix for the two biggest. A board that shows only the 96.2% invites the question of where the other 3.8% went.

Can the capstone poster be printed for an expo stand?

Yes. It is A0 portrait, 841 by 1189 millimetres, and exports as PDF, PNG or JPG at twice its working canvas, 3,312 by 4,680 pixels, about 100 dots per inch at full size. There is no bleed, crop marks or CMYK, so check the print shop takes an RGB file.

What should a capstone poster say about cost?

Give the parts cost per unit and what it buys, beside the brief that set the budget. This team had a ceiling of $40 and built the device for $38, against smart meters that cost hundreds of dollars a home to fit. A judge from industry reads the cost before the accuracy.

Build your own in about a minute

The button below opens the generator with this use case already described. Change the wording to match your own, generate, then edit anything you like.

Make one like this

Other poster examples

Simulation study poster example

A simulation study poster must say what the model assumes as clearly as what it found. This A0 landscape example runs an agent based evacuation model of a 1,200 seat hall under four exit layouts, 500 runs each. Adding a fourth exit at the rear cuts mean clearance from 212 to 168 seconds, 21% faster, and the 95th percentile from 251 to 192. Four model limits sit right beside that headline.

Controlled experiment poster example

A controlled experiment poster has to show that one thing changed and everything else was held fixed. This A0 portrait example grows 288 lettuce plants in 12 trays under three LED spectra for five weeks in one chamber. From the same 0.8 g start, the red heavy spectrum ends at 118 g a plant against 96 g under balanced white, 23% more.

Qualitative study poster example

A qualitative study poster has to let the participants speak in a few words and still show how far each theme reached. This A0 portrait example reports 36 interviews with unpaid family carers about getting respite: four themes, each with one quote under 30 words and a count, led by respite as a fight in 28 of 36 interviews.

Want the steps in the builder? Read Add charts and diagrams, then choose the template, theme and size. For everything this generator can do, see the poster maker.

Sources

Written and checked by the OneCraft team. Last checked .