Measure distance of 1.5-2m, accurate to within +/- 2mm

Hi guys,

To take the measurements my plan was to use TOF sensors

I have tested VL530x and VL531x sensors, but they are not very accurate. Here is a list of things I have tried, in order to improve accuracy:

-SPAD Region of interest: 4x4, 8x8, 16x16

-reflective surface: Gloss white, reflective tape, bicycle reflectors, aluminum foil.

-reflector size: 90x40mm, 120x120mm

-ambient lighting: high and low

-software changes:
distance mode: high, medium, low.
Timing budget: various settings between 50ms and 500ms.
Result filtering: exponential filter, running average, and both together.
Measurement modes: continuous, single shot.

Doesn’t matter what I do, I can’t get any better than +/- 5-10mm

Does anyone have any tips on how to improve the accuracy of these sensors, or know of some other sensors that are more accurate?


Is this device intended for measuring in the depot and not live out on the trace track?
If so, I suggest a well calibrated rod and a caliper/micrometer arrangement.

I have a Leica Disto D1 laser-distance measuring device

It has a bluetooth-interface and a smartphone app.
There is a continously measuring-mode that does a measuring once every second

not cheap as you would need two of them and I guess two android-devices to use the App but accurate

same price-range is this one

best regards Stefan

I'm used to railroad tracks. We use like a scale with a moveable caliper like indicator telling, in millimeters, how far, plus or minus, we are from nominal.

A well calibrated rod equipped with a linear potentiometer and a sensing tip, all fed into an Arduino looks doable to me. Calibration is needed (once now and then) for every option. Such a linear pot likely have a range of max some 50 mm, 2". Prepared, known, checked, steady positions for the pot to cover the total span, 1.5 to 2 meter.... Sounds doable and reliable to me.

I would imagine you will need some way to measure the steering is straight ahead with reference to the rear.
Otherwise measuring just these two as you say would be useless due to the geometry of the front steering arrangement.
Perhaps reference to the pitman arm if conventional.

You can't be the first in the world to want this, so the coolest thing you can do is find out what the others are doing. If you really need to do this, you may well find your best answer in reply #2

As mentioned previously, as you move the wheels from their centre position, each moves a different amount due to the steering geometry.
Therefore it would be essential to find this central position and lock it in place, otherwise the toe in reference is no longer viable.
Previous answers #2 and #8 might well be a good starting point.

There are more accurate TOF sensors but, since they're not hobby grade the price goes up.

Wild idea.

After you've searched on the internet with words such as "arduino machine learning edge detection" and, from your reading decided its worth the learning curve, I am sure you'll get a good working product.

I was just able to get a amg8833, a very small thermal camera with only 8X8 pixels, using a ESP32 to distinguish squirrel, duck, person and upon person detection generate an action signal using machine learning (ML).

I got some data to look at this morning, from the project, then I am going to test out the code I wrote to combine 2 images from 2 amg833's.

Oh, almost forgot. Google has one of the largest open source stores of tensors (trained machine learned models) around. The trick to getting access to it is that you must contribute to the project of tensor collecting. I mention that because there is most likely a trained model for a properly aligned car tire in their store.

When you do this doesn’t the car body work get in the way of any light beam going across the car ( as you need to be half way up the wheel - I can’t imagine any car where you can see straight across unless you make some holes ! ) ?

Also the accuracy you want is not good enough , with toe in typically being a couple mm , knowing that to +- 2mm is useless .
Buying a tracking gauge and be done with it

One way to get around the accuracy/precision issue with the TOF detectors (or any detector) is based on the statistics of confidence limits and confidence intervals. For instance, assuming a standard normal distribution curve for the data, you can calculate that if you have a standard deviation of 7 mm, and desire to know that a value is 10 +/-1mm, i.e., in the range 9 to 11 mm, 99% of the time, you would have to take at least 325 measurements to ensure the accuracy is within the desired range. At 0.1 mm the number of observations goes up to over 32K observations.

Gave you a like for following up with your solution.

1 Like

This topic was automatically closed 180 days after the last reply. New replies are no longer allowed.