loading
 
People tracking in street within zones
Guy from Netherlands  [2 posts]
3 years
I want to do some people tracking on a publicly available webcam. If people walk in certain areas I want Roborealm to send a keyboard press that corresponds with that area (see example picture).

The webcam source is a YouTube live stream that I capture with the screencapture module.

I've tried using the movement detection module but I'm not sure how to identify the zones, how to define what should count as a person walking through and how to let it send the keyboard key presses.

Finally if this would run 24/7 I would probably need to do some tweaking for changing light conditions. Maybe shut it off during certain time periods.

If someone has a .robo file that kinda does this or maybe a tutorial or tips that would be great! I think this software should be up to the job, I just need to find out how.

 
Steven Gentner from United States  [1370 posts] 3 years
Guy,

There isn't a robofile that would just do the job as each of these require different configurations/approaches depending on the angle of the image.

While yours isn't ideal you can still get some useful information out of this. It will NOT be 100% accurate as we're going to make some assumptions that may or may not hold true depending on the actual person walking through and the number of people walking together.

1. The idea is to use the movement module with a long background (1000 frames should do it). That will slowly adjust for overall lighting as the day progresses.

2. Using this the movement module will indicate local differences between the background and the current foreground. Setting the movement to White will create a mask that can be quantified.

3. Now you need to isolate the areas you are interested in. Best to do this one at a time. You can use a mask (i.e. white image) (Load_Image, Math module) or the Crop module to isolate only those areas you want to count.

4. Finally, and this is the tricky step, you want to count the number of blobs within the interest area to determine how many people are there. The trick is that you will have blobs that are not part of a person and blobs that are combinations of people. Its easy to eliminate the smaller ones but the number that constitutes a blob contains more than one person is difficult. As the system is not identifying people but rather groups of pixels that are not similar to the background, this number will depend entirely on the image. This is where the inaccuracy comes in.

Because this is a bit tricky to setup, can you post a couple seconds of that video somewhere we can download and give you a head start? Ideally, a couple people would be walking through which would give a small sample on which to base that count.

Thanks,
STeven.

This forum thread has been closed due to inactivity (more than 4 months) or number of replies (more than 50 messages). Please start a New Post and enter a new forum thread with the appropriate title.

 New Post   Forum Index