Following a person, is TSL1401 the thing to use?

Wolfbrother

Hi all,

I'm doing a new project where I am hoping to have a mounted head follow a person as they walk around it, until they get to the max rotation allowed and then the head would swivel around to the other side to continue watching the person. Like you would do if·you were seated, etc.. I'd like to kind of have a good shot at recognizing a person out of maybe several, so I am thinking to use the TSL1401 and have it look for the same color blob, hoping that people aren't all wearing the same clothes.

My questions are the data brief says it focuses 25mm to infinity, so I'm guessing people 10 feet away or so is fine, is that correct? Does this seem like the best way to do this or is there an easier method yet?

Thanks,

Dave

Phil Pilgrim (PhiPi)

Dave,

You would probably be disappointed with the TSL1401 for an application like this. First, it is not a color sensor. So recognizing a particular color blob would not be possible. Second, it sees in one dimension only. So, with the sensor axis oriented horizontally, you would see a horizontal slice of a normal 2D field of view. This slice might include a person's belt, their shoulders, or their eyes and the bridge of their nose.

Now, having said that, if you were tracking a single person, silhouetted against a bright background (i.e. backlit) the TSL1401 would give you at least half a chance of following their motion. What makes this work is a single, high-contrast subject that's not mixed in with other visual clutter.

-Phil

Wolfbrother

Hi Phil,

I was kind of hoping to take advantage of·the sensor·just doing a slice by the orientation. If this thing was located about 32-36 inches from the ground, I think that's close to the center of mass of most people. I can control the enviroment around the sensor, since it's a gallery setting it's almost always the case those walls are white. I misspoke when I said color, I was thinking contrast, since I saw the forum posting where you scanned the currency.

So, if this isn't the technique to use, does anyone have a suggestion I should try? I have also thought about a CMOS imager mounted in the ceiling, tracking the people as they move around this thing and then bringing this down to the head. Maybe that's better, but could wind up in some wire routing nightmares. Maybe the CMUCAM is my best plan?

Thanks,

Dave

David B

I've had an idea in the back of my mind for doing something like this for a Halloween stunt that I've never tried, but maybe it would work.

If you had two microphones in the ears of the head, you could measure the microphone readings and either from timing of specific received waveforms, or just from comparing the overall sound power of the two microphones, you could swivel the head to equalize the received power or to equalize the timing of the received waves.

If it worked, then you could even have the eyeballs swivel in rapid response to sounds, and have the head slowly follow. That way the head could follow a two-way conversation between two nearby people. Creepy!

Wolfbrother

Hi David B

That's a good idea, but I have to plan for people being silent. How people equate museums and galleries with libraries is beyond me, but they can be as quiet as little mice.

I'm going to order a CMUCAM and make that do what I want. I would have liked a cheaper solution though.

David B

Here's another idea.

A few weeks ago I bought the Parallax IR sensor. I only just got it running late last night, so I have a grand total of about ten minutes of experimenting with it, but the results so far are very encouraging.

With the sensor sitting on my desk, pointing up, and a two-line LCD displaying the sensed value, I could wave my hand two feet in the air above the IR sensor and see an immediate response. I got the ten degree model, so its beam of detection is pretty tight, and it showed as I moved my hand around over it.

Suppose you were to rotate this sensor like the beam of a lighthouse, scanning the full 360 degrees of the room. With a ten degree sensor you'd have about 36 segments of separation that you could track and monitor. The location of people would show up as peaks in the scan. If they were walking, you should be able to track their movement quite well, I would guess.

Maybe others with more experience with this sensor, like Phil, would know better than me how suitable this sensor might be for this task, but so far it looks to me that it could work.