people detection

[mfrz]

hello,,

I am using the quadcopter to detect object such as people.
I already add vision sensor to be used to detect object.
I think it is a proper way to do it.
However, I don't know how to start to code it so when my quadcopter move above people, it can detect the object (people).
Of course, it will detect only the head vertically.
I already look at the demo example Scenes/blobDetectionWithPickAndPlace.ttt but I still don't understand it.


Can anyone guide me for starting code part?

Thanks.

[DannyYeong]

Same question here, was going to post up something similar. However I'm using my basic self-made quad which only covers translational movements (no rotations balance like the provided one in VREP).

I was thinking using of path planning the quad to the human and then once detected, other quads within vicinity will converge to the same target. But currently I'm not sure if the path planning should be directed to the human OR let the Quad sweep an area and then program it to move towards human (ignore default path plan) after detection.

Regards,
Danny

[coppelia]

Hello,

you have several ways of doing this. You could use a proximity sensor, or a vision sensor. The close and far range of the sensors need to be carefully chosen, in order not to see the floor. Or, if you want to make it more realistic, filter out the pixels read by the vision sensor (e.g. based on color, on the depth map, and/or blob detection).

But much simpler that this is if your vision sensor is fixed to the terrain like in this example, where everything above the terrain is detected as a pixel in the vision sensor image.

Cheers

[mfrz]

Same question here, was going to post up something similar. However I'm using my basic self-made quad which only covers translational movements (no rotations balance like the provided one in VREP).

I was thinking using of path planning the quad to the human and then once detected, other quads within vicinity will converge to the same target. But currently I'm not sure if the path planning should be directed to the human OR let the Quad sweep an area and then program it to move towards human (ignore default path plan) after detection.

Regards,
Danny

Thanks Danny for your great idea. However, I am thinking that the quad's movement will be random and once it detects people, it will move toward people and take a picture of that people as an image to be stored. So far, I want to focus on detecting people.

Hello,

you have several ways of doing this. You could use a proximity sensor, or a vision sensor. The close and far range of the sensors need to be carefully chosen, in order not to see the floor. Or, if you want to make it more realistic, filter out the pixels read by the vision sensor (e.g. based on color, on the depth map, and/or blob detection).

But much simpler that this is if your vision sensor is fixed to the terrain like in this example, where everything above the terrain is detected as a pixel in the vision sensor image.

Cheers

Thanks coppelia,
I am thinking to use vision sensor since I assume as the easiest way.
For now, I am trying to understand the example that you mentioned.

So, my main idea is when the quadcopter detects people, it will move toward it and take a picture of it and then store it.
Is it achievable by using vision sensor to detect and take a picture?

[coppelia]

Yes, you can do that. Make sure to look at the various code samples, models and scenes that come with V-REP.

Cheers

[mfrz]

Can you suggest an example to me that can guide me?

Thanks.

[demircik]

I want to know which sensor you are using as a vision sensor. Is it any optical sensor or any other?
If you are using any optical sensor then ofcourse there would be lot of problems while detecting people.
If you can use image processing then it can provide you more efficiency, but it would require much complex circuitry.

printed circuits assembly

[coppelia]

Yes, a vision sensor is a optical sensor, or a camera sensor. And it is correct that it often requires some complex image processing algorithms to extract useful data from a vision sensor.

Cheers