Which is better for depth applications: D415 or D435?

[ASriv7]

Which is better for depth applications: D415 or D435?

[MartyG · Accepted solution]

They are both excellent. The D435 benefits from a faster shutter and larger FOV. In regard to depth noise, the D415 will typically have less noise though. Information in the link below explains why.

[WTatt]

It really depends on your application - both devices have their strengths - and weaknesses.

It is very important to understand however - that if accuracy of the depth data at distances beyond about 100cm is important to you then the D435 is not a good match.

The D435 is good at capturing a wide field of view - but the "quality" of the resulting depth data degrades significantly the further you move away from the sensor.

As such - near field tracking is better than distance tracking.

The D415 has a narrower field of view - but the quality of the data is maybe 2x that of the D435 for the given point in space.

Westa

[RTasa]

Have you tried measuring beyond 10 meters for accuracy?

[WTatt]

For those distances - it would really depend on what you mean in terms of accuracy - but again the d415 would be our pick for that sort of range.

Depth errors - which are a fact of life with the types of sensors being used ( and appear like dimples if the depth data is projected on a plane) scale at a rate equivalent to the square of the distance away from the sensor.

And the 435 scales at a rate in excess of double that of the 415. If you look at the charts provided by Intel you can see how rapidly the depth errors increase.

At 1 meter the errors area round 3-4mm for a 435 and 1-2mm for a 415

And at 5 meters this means something like a potential depth error in excess of 120mm when using the 435 - while the 415 sits around 30mm-35mm

Out at 10 meters - which yes the sensors can see to infinity - the level of errors for d435 would be off the top of the graph

There are ways to mitigate these values somewhat by adjusting the disparity shift setting - but regardless - it still would result in significantly greater accuracy from a 415.

Westa

[RTasa]

reasonable error would be in meters at 50 meters.

I noticed the 415 has increased pixel density which the 435 does not.

415 is 1080 HD while the 435 is only 720p.

This does make a difference as to how accurate it can be looking at the distance.

May have to dump my 435 for the 415 as I am trying to see distant objects.

Maybe a dual 4K camera with a difference engine built in would be good and a way to align it with a lower res

VGA image for inference would make an interesting camera.

[RTasa]

After seeing the graph it looks like the max distance is about 14 meters with a ~1 meter variance.

13 meters is 39 feet or 2.5 car lengths.

Bummer...

[MartyG]

In explaining the link between vision distance and disparity shift, the other, prettier version of the tuning document says on page 15 that stereo cameras will see infinitely far so long as disparity shift = 0.

[jb455]

I did some (unscientific) testing at long range with the D415 and D435, measuring 3m, 6m & 9m distances perpendicular to the camera at 5m intervals of range.

With the 415 I found the error to be <10% up to 20m, then stepping up to around 25% up to 40m then sharply going up to 100-150% up to 50m (I didn't test further than that).

With the 435, error was <10% up to 15m then became much more variable. I was able to get low errors further than that, but it was more the exception than the rule.

At long range, I found that one of the big limiting factors was resolution of the colour and depth images. (I have depth aligned to colour to measure on). When you're so far back, each pixel covers quite a large area so it's hard to accurately choose the correct points to measure on the colour image, before even worrying about the quality of the depth data behind it.

I do remember reading about a D450 or something which Intel had made with a larger baseline specifically for long range applications; it may be worth you contacting Intel for information on that (https://realsensesupport.intel.com/hc/en-us https://realsensesupport.intel.com/hc/en-us is probably the best way to get a response)

[MartyG]

Regarding the D450, what I have learned is that until a product appears in the Click store, it is best to regard it as not existing. This is because in-development projects can be subject to cancellation before they are released (like with the RealSense Smartphone Development Kit and the Project Alloy headset) or having their technology merged into other projects.

[WTatt]

HI Bob,

These are similar results to those we have experienced and witnessed in our testing - THERE ARE WAYS to improve the distance sensitivity - using the disparity setting - but this comes at the cost of loosing the foreground details.

Westa