Creating a 3D Linear Motion Detection System for Scientists

[Lagomorph]

Hi,
I need to make a 3d linear motion detection system for, oh... say a ball. I was thinking of setting up 3 small base stations that would ping something inside the ball and would be wired to a microcontroller that would use a triangulation algorithm to figure out the displacement of the ball along the 3 axes.

I know that this isn't a new problem and that the devices to do this should exist.

- Does anyone out there know what devices I should be looking for to construct this thing?
- Does anyone know if there is already an existing triangulation algorithm for such a task?
- Any other comments / suggestions?

I'll take any help I can get. I have incredibly stringent deadlines. Thanks ahead of time for the help.

 

[berkeman]

The best solution depends on lots of things, including cost constraints and how the ball is used/accessed.

-- Video cameras work well, as long as the ball is not shadowed (like they are good if the ball is just bouncing around on its own), but obviously are expensive.

-- If you can connect 3 shafts to the ball orthogonally, then you can use linear encoders to record the delta-movements of the ball.

-- If the ball is isolated and not shadowed, you could probably use 3 coordinated ultrasonic transducers to ping its position...

-- If you can put an ultrasonic pinger thing inside the ball, you could use 3 ultrasonic pickups to triangulate on the pings (but this puts some constraints on the other objects that may be in the room with the ball).

More details on the constraints would be helpful.

 

[Lagomorph]

Right, I guess I wasn't very specific on the use.

I would like the ball to be hand-held and freely moved in space. Therefore, I don't think I can use anything that requires line of sight, as I can't guarantee that the body will not be blocking that line of sight. I was originally thinking of using accelerometers and a gyroscope, but that can get fairly expensive and perhaps more complicated than the solution I mentioned above. Cost is of course an issue, but not the biggest one.

I was thinking about the ultrasound idea. I figure anything that transmits a high frequency omnidirectional signal should work ok, wouldn't it? Would ultrasound work well for this purpose? Are there any commonly used parts to perform this pinging? I was first thinking of having the ball contain something like an ultrasonic pulse generator that the base stations would pick up, but perhaps it would work better if the base stations would emit the pulse and the element in the ball reflect it. That way, the stations are less dependent on the ball, which might be a good thing.

 

[berkeman]

By hand-held, do you mean that a person can block the line of sight (or ultrasound path)? Or will the person always be to one side, so that 3 detectors in front can have a straight shot always? As for the ultrasound angle, it would be much more efficient if the ball can generate the pings. Quiz question -- why?

 

[Lagomorph]

Since the ball should be handled freely in free space, it is very much conceivable that the user could get between the ball and the base stations. Would ultrasound not penetrate the human body? How about an RF signal?

I figure if the ball is emitting the signal, we don't need anything else sent or reflected, which would be more efficient. On the other hand, that would probably mean more parts inside the ball and more power consumption within the ball, which I suppose could be inconsequential.

So do you think that ultrasound would work ok, or do we need something that more easily penetrates solid surfaces?

By the way, I've been trying to find existing technology that would perform any of the tasks that I need and I thought that proximity sensors might be a solution, but I can't find a single one that can sense farther than about 3". Am I looking in the wrong places?

 

[berkeman]

Both RF and ultrasound will be distorted by the presence of the body. Longer wavelengths of RF will be affected less, but would not generally be practical to emit from a small ball with internal antenna (especially isotropically...). Ultrasound or even sound (a beeping) has the advantage of a much slower propagation velocity, so the circuitry to triangulate on the signal is easier to build.

Again, going with the theme that longer wavelengths are less distorted by the presence of the body, could you use an audible beeping from the ball for it's position sensing? Or would that be too obnoxious to the user? Also, keep in mind that you'd want to minimize other reflections in whatever technique you were using, so for example for the ultrasound example, you'd like soft walls in the room or cubicle where the device was being used.

 

[berkeman]

How about a different idea entirely. Put a position sensing arrangement on the person's hand that is holding the ball. Then just sense hand movements (with rods and linear transducers, for example).

 

[Lagomorph]

Do you really think that they ultrasound and RF will be distorted that badly? Wouldn't that mean that cell phones would would work very poorly if you stood with your back to the base station? I think beeping would be extremely annoying and render the device unusable. Plus, the pulses would need to be emitted at a high frequency for enough sensitivity to make a fine position grid. I don't know if that would be a problem in terms of pure technical implementation.

What about bluetooth? That's what the Playstation 3 controller uses and I know for a fact that it passes through bodies, walls and any other objects regularly found in a room with absolutely no noticible performance changes.

The first thing that I considered when approaching this problem was accelerometers. Unfortunately, because of the control scheme that I want, rotational displacement matters, as it affects the direction of the acceleration that is sensed. After looking into tri-axis gyroscopes, I figured that the task would be altogether too costly and probably more difficult than the pinging sensor solution.

By the way, thank you very much for the discussion.

 

[berkeman]

Yes, for your position sensing purpose, the distortions from the body would be a big hinderance for ultrasound or RF. For position sensing, it is more of a radar-like sensing, with time of flight the key variable to the 3 sensors. You want a sharp wavelet attack and decay, so that the sensors can be used to detect the difference in arrival time of the 3 paths. Multi-path is a big problem for this kind of setup.

Bluetooth and cell phones and other RF technologies have to deal with multipath issues, although mostly in the reception amplitude null problem, not time of flight problem. The modulation length of the RF signals is long compared to the human body, for example, but the carrier wavelength is not. So the information is not typically scrambled due to local multipath issues, but the amplitude of the received carrier signal can null quite badly when you are in just the right (wrong) spot where multi-path causes a big null.

Back to your project -- what about the idea of having a glove or other arrangement where you sense the position of the user's hand instead?

 

[Lagomorph]

Ok, so no go on the RF or ultrasound sensing unless I want to impose the line of sight constraint. Better to find that out sooner rather than later.

Whether I'm sensing the position of the hand or a hand-held object, I'm not sure what kind of solution you're proposing. As I mentioned above, using 3 linear devices to sense 3D motion requires compensation for rotational motion. Is there any way to do this aside from tri-axis gyroscopes? If I do use something like a tri-axis accelerometer and gyroscope combination, that would require many more parts within the actual controller, a ton more microcontroller programming to rotate the axes based on the gyro output and I would need to transmit the signal wirelessly to a receiver. I couldn't expect the user to keep his/her hand in the same position, so the gyroscope would be necessary. From what I found, those tend to get pretty expensive.

Is that kind of sensing system what you were talking about, or did I misunderstand? If I didn't, am I missing something that could make it not extremely expensive and difficult?

 

[berkeman]

Position sensing is something I've spent a lot of time working on -- my first and best patent was in that area, and focused on the lowest-cost, high-resolution position sensing possible for a particular situation.

From what I've gleaned from your posts so far, my solution would be to sense the hand position, not the ball position. What fundamental limitations would happen because of that? (I'm not pushing back hard here -- just being a devil's advocate in your brainstorming process...)

 

[berkeman]

Position sensing is something I've spent a lot of time working on -- my first and best patent was in that area, and focused on the lowest-cost, high-resolution position sensing possible for a particular situation.

http://patft.uspto.gov/netacgi/nph-...&OS=berke+AND+graphics&RS=(berke+AND+graphics)

 

[Lagomorph]

What would be the difference in sensing the hand rather than the hand-held device? Both would be essentially in the same position at all the same times.

Regardless, I need to fulfill some stringent deadlines, so for now I will impose the constraint that the user must keep the ball in line of sight of 4 sensors and use the pinging idea. It requires less hardware to implement than any other solution I can think of, which is preferable at this point.

The only way that I can think of detecting distance from a base station is to have the ball emit an RF pulse and an ultrasonic pulse at the same time, have the station detect both and use the difference in the arrival times of the two pulses to determine the distance of the ball from each of the four sensors. I will place the 4 sensors in a rigid pyramid construction, so I will know the distance and position of all the sensors in relation to each other. I figure this would allow me to calculate the position of the ball.

Does that seem like a good solution? If so, do you have any idea what kinds of frequencies I should be working with? The biggest problems I am having are a lack of real-world experience with available parts and standards and a lack of advising. Would you happen to have any idea what I should be looking for in terms of parts?

 

[es1]

Some questions:

What will be the ambient environment? i.e. a baseball park, or a controlled lab.
Does it have to be real time?
How well do you have to determine the ball's position?

Personally I think Berkeman's suggestion is the best:

-- Video cameras work well, as long as the ball is not shadowed (like they are good if the ball is just bouncing around on its own), but obviously are expensive.

But cut the cost by renting (or ask a friend to borrow) a digital camera if you don't already own one. Then use existing software libraries to assist with the post process (break apart frames, create bit maps). Of course this assumes the background is easy to filter out and reference distances can be determined.

I think writing post processing software would be way easier than any kind of trianglation method where you have to build the sensors (unless you have resources to burn). I played with a similar thing in college, just trying to determine the static position of a light blub in an otherwise empty dark room, and was plagued with accuracy issues. In the end I was accurate to about a 50cm diameter after calibration of the analog circuit (an amplifiers class project). But it was only a (part time) week's worth of work.

 

[berkeman]

The difference between sensing the hand versus the ball is whether you have to have a general ball that can be used, versus a tethered or instrumented ball. If you have to be able to sense various balls like a baseball, tennis ball, golf ball, etc. all being moved around in your apparatus, then instrument the hand. If it's always the same ball, instrument the ball.

I don't think that combining RF and ultrasound buys you anything. Just use ultrasound pings and use the delta-T of the time of arrival at 3 receivers to triangulate on the emitter position. Dealing with ultrasound is an order of magnitude simpler than dealing with RF for this project, IMO.

The simplest position sensor still would be to have a ball that has 3 rods coming out of it that go through 3 linear position sensors that are placed in front of the user. Can you make that setup work for your project?. It would be a very practical 3-D position sensor arrangement. You could use optical pickups (with stripes on the rods) or little wheels with rotary encoders as the pickups for the linear displacement sensors. Oh, and you'll want some "home position" somewhere that you can put the ball into to reset the overall motion sensor.

 

[iconoclastic]

Position Detection Problem

Hi,

I have a similar problem. I am trying to determine the 3d position of a point in a space. Assume that the point to be tracked is the tip of a pen. I would want to determine the x,y,z of the tip with resolution of 0.1mm. What would be the way to approach this problem ?

 

[berkeman]

Hi,

I have a similar problem. I am trying to determine the 3d position of a point in a space. Assume that the point to be tracked is the tip of a pen. I would want to determine the x,y,z of the tip with resolution of 0.1mm. What would be the way to approach this problem ?

What are the physical constraints of your situation? What do you think about the ideas expressed so far in this thread? Can you use any of them? What other techniques have you read about so far -- are any of them close to what you want?

 

[iconoclastic]

Hi,
Thanks berkeman for your reply.

The problem i am facing is not just linear motion detection. In fact it is the determination of 3 degrees of freedom of an object in space, rotation/tilt in two axis(say, Z and Y-axis) and translation motion in another axis(say X-axis).

Using cameras and Digital Signal Processing stuff, costly solution for me.
Used Optical motion tracking devices, again costly. I am looking at an economical solution.

I have tried out a few kinematic mechanisms and used encoders to detect the rotations and stuff like that, but, i am not able to figure out a way to determine the linear translation of the tool tip of the pen.

The degrees of freedom i need to detect are tilt about two of the axes and rotation about another axis.

 

[berkeman]

Well, if you can tether the object with three rods that slip through a yoke at the object, and if you can make that yoke so it can rotate in the ways you want, then you can use 3 linear encoders on the rods and rotary encoders on the rotating parts at the yoke, to give you all the data you want. You can just use reflective optical photodectors and patterns on the rods and rotating things to give you the quadrature pulses you need to detect movement. You will also want a "home position" somewhere with a switch, to get you started off with an absolute position that then gets updated with the relative motions detected by your encoders.