Slip Angle Sensor - General Idea

[Caito]

Hi Guys!

I was thinking if there was a way to make a fairly "cheap" slip angle sensor. Or put in other ways, a device that could measure longitudinal and lateral speed with respect to ground. Based on other commercial sensor I've seen, I believe ca. 25Hz is enough.

Of course, one alternative is to use optical mouse sensors. Although they have a limited speed capability, you could put another sensor to have a broader speed range. Still, it doesn't sound easy enough, you need optics, sounds unreliable (at least for me).

Nowadays having so much data throughput and processing capability I thought why don't just point a camera to the ground. Then you just find a reference in the image and see where it moved on the next image. It needs quite some processing, but that's offline and is "easily" done. Basically you can move images to find the point of max correlation, or actually search for on piece image on the other, etc. There are many ways to do this.

My question is how blurry would that image be?

As some ball park number. Imagine capturing a spot of 20x20cm and a frame of 500x500 pixel, this means 40mm per pixel. If you want a "still" frame, you would want less than a 40mm movement. Suppose we want this until 50 m/s (180 kph), this would be a 1/1250 shutter speed. With 20cm movement being the maximum allowed between frames, we need a 250Hz frame rate. As you increase spot size you reduce the frame rate, but would increase error I guess.

But I don't know much about photography so that might as well be pure bs.

Do you think something like this could work? Is it worth a try? What problems can you see?

[langwadt]

Hi Guys!

I was thinking if there was a way to make a fairly "cheap" slip angle sensor. Or put in other ways, a device that could measure longitudinal and lateral speed with respect to ground. Based on other commercial sensor I've seen, I believe ca. 25Hz is enough.

Of course, one alternative is to use optical mouse sensors. Although they have a limited speed capability, you could put another sensor to have a broader speed range. Still, it doesn't sound easy enough, you need optics, sounds unreliable (at least for me).

Nowadays having so much data throughput and processing capability I thought why don't just point a camera to the ground. Then you just find a reference in the image and see where it moved on the next image. It needs quite some processing, but that's offline and is "easily" done. Basically you can move images to find the point of max correlation, or actually search for on piece image on the other, etc. There are many ways to do this.

My question is how blurry would that image be?

As some ball park number. Imagine capturing a spot of 20x20cm and a frame of 500x500 pixel, this means 40mm per pixel. If you want a "still" frame, you would want less than a 40mm movement. Suppose we want this until 50 m/s (180 kph), this would be a 1/1250 shutter speed. With 20cm movement being the maximum allowed between frames, we need a 250Hz frame rate. As you increase spot size you reduce the frame rate, but would increase error I guess.

But I don't know much about photography so that might as well be pure bs.

Do you think something like this could work? Is it worth a try? What problems can you see?

an optical mouse is a camera

[Caito]

an optical mouse is a camera

Yes it is, but if you look at the sensors, they're usually tiny and very specific. Quoting one of Agilent's datasheet, max speed is 12ipm (roughly 110kph) and would require specific lenses to make it work at the distances and speeds required. It's not simple enough.

[NL_Fer]

The cars have g-sensors and gps right?

[mrluke]

Thinking out loud.

Do you really want a super sharp image?

Using a known shutter speed wouldnt the extent and direction of the blur tell you the movement / slip? Actually if you could measure it very precisely wouldn't it actually give the slip angle?

[Tim.Wright]

Cheap slip sensor? Rig up a caster wheel with a measurement potentiometer. The caster wheel will (more or less) follow the road - therefore the angle between the caster wheel and the vehicle body is your slip angle.

Another method is to calculate it from a lat acc and yawrate sensor. It requires an integration so it will drift over time but its useful for short maneuvers like step steers etc.

Depends what you need the data for in the end?

The cars have g-sensors and gps right?

That's not enough to calculate slip angle. GPS is horribly noisy and imprecise. You need accelerometer AND gyroscope measurements to correct it and even then its not a particularly robust or precise measurement.

[Greg Locock]

However I think the camera solution has legs. If I were to do it i wouldn't get too caught up with the optical side (a statement I will likely regret), and basically worry far more about the processing side. If you imagine two consecutive frames of a speckled pattern, how quickly can you decide on the 3 possible transformations between the two? (rotation and x and y).

As Tim says, measuring it more than one way is probably necessary. We use differential gps instead of gyros these days, I don't know how much our accuracy has suffered.

[Tim.Wright]

I've experienced, and have heard other stories of problems with even high-end (50k€+) inertial+gps slip measurements. I've had signal drift down straight sections of track and I've heard issues of them exhibiting different signal offsets depending on which direction you drive. Plus numerous reliability problems. So Ive grown pretty skeptical of these systems.

I'm currently of the opinion that if you want to measure the road velocity - you should take a direct measurement ot the road.

[Shooty81]

If you imagine two consecutive frames of a speckled pattern, how quickly can you decide on the 3 possible transformations between the two? (rotation and x and y).

What if you use the non-zero exposure time of a single frame and calculate the angle directly out of the streaks of the asphalt pattern of one single picture?

[notsofast]

Do you really want a super sharp image?

Using a known shutter speed wouldnt the extent and direction of the blur tell you the movement / slip? Actually if you could measure it very precisely wouldn't it actually give the slip angle?

What if you use the non-zero exposure time of a single frame and calculate the angle directly out of the streaks of the asphalt pattern of one single picture?

This idea (from both of you) seems to be the most straightforward. As long as the asphalt is not too smooth. I guess you would want to aim the camera somewhere between the wheels, away from the tire tracks that are rubbered in.

[Greg Locock]

You'd have to write a function that'd identify streakiness..

You are in luck, the following course includes enough image processing to get you started, it is free, not too hard, and I thought it was fun. Plus you get a limited use edition of the latest matlab, which makes me annoyed as I'm stuck with 2009a(more accurately my PC is set up and works, stuffing with matlab versions is a guaranteed way to have to sort out the other programs that use it),

https://iversity.org/en/my/courses/mode ... ouncements

Incidentally I was right to be cautious of my gung-ho attitude to the photgraphic side of things. At 25 Hz I will need a field of view of 100/33.6/25= 1.11 m at 100 kph, bit tricky from underneath a car. My guess is they run a much higher frame rate.

I have a sneaking suspicion that running a camera at each end of the car will give much better results than trying to do everything off one camera, that way you'll see a difference in lateral velocities, rather than a lateral velocity and a slip angle.

[Caito]

I came across this sensor, that might be almost all needed to implement an idea like this one.

https://pixhawk.org/modules/px4flow

Cheers,
Andrés

[Greg Locock]

Excellent. (Why can't i mod you up? Android.) well now I have

[Brian Coat]

Raspberry Pi + Picam?
https://www.raspberrypi.org/blog/vector ... stimation/

I must admit "Raspberry Pi!" is my kneejerk response to just about any custom digital look-see project.

But why not? Cheap as chips and what ever you want to do, someone probably already did 80+% of it for you because there is such a high nerd-base using them.

[graham.reeds]

What about gps and a 9dof imu? The gps gives updates 5hz with interpolation from the imu.

The imu will not only give you everything you need so I can't understand why you are thinking of turning the car into a giant wireless mouse...