Hand tracking output problem

ELAN recognizers
1

Hello everyone,

I’m using the SKIGUI and the handtracking recognizer included with ELAN.

I know this is a program in developing phase but I’ve found a recurrent problem with the program.

Once the parameters are set with the SKINGUI; I proceed to do the handtracking with the ELAN extension.

The csv file containing the coordinates of each hand and the head on each frame of the video is incomplete. This file jumps from the frame 100 to 201 so I don’t have access to the coordinates of the hands and head during this period. The video created by ELAN containing the ellipses marking the hands and the head is complete and has no problem during this period.

Is this a known problem? If so is there any solution?

Let me know if you need any other information or need further explanation.

Thanks a lot for your help and time,

Luis Morís.

2

Dear Luis,

unfortunately this is the first time I hear of this error. Does the video behave normally between frame 100 and 201 (that is, between second 4 and 8)? Camera motion, changes in lighting, tracker not working, anything? Does

Maybe with some more details I will hopefully be able to reproduce the error.

Thanks in advance,
Stefano

3

Dear Luis,

Could you confirm if in the video between frame 100 and 201 the ellipses are detected correctly as the hands? Then they will have some colors assigned. If they appear black in the video, it means they are not recognized correctly and that’s why you don’t see any data in the CSV file.

Best,
Przemek

4

Hi Stefano,

The video has no problem between 100 and 201, even more the video containing the ellipses from ELAN is fine, and contains the ellipses at all moments.

I’ve shared the video I used for tracking the hands (prueba 10.mpg), the SKINGUI parameter file(Skin_final2.xml) the csv output (Prueba_10_22_April2.xml) and the video output of ELAN (10.mpg) so you can reproduce the error.

I’ve tried with two different videos and in both the problem appears.

The parameters I’ve used for the tracking in ELAN are:

Under the tab Video recognizer -> NEW Human motion analysis and annotation:

Advanced:
Speed threshold…: High
Background image…: true
specifies whether analyze…: true
update_rate: 100 (What’s the purpose of this rate by the way?)
Threshold used to decide…: 40

Settings:
Adjust amount of log messages: verbose
Resolution of output video…: full

xml containing skin colour…:
File->The one in dropbox

Here are the links:

https://dl.dropboxusercontent.com/u/60241410/Prueba10_22_April_2.xml
https://dl.dropboxusercontent.com/u/60241410/prueba%2010.mpg
https://dl.dropboxusercontent.com/u/60241410/Skin_Final2.xml
https://dl.dropboxusercontent.com/u/60241410/test10.mpg

Thanks,

Luis

5

Hi Przemek,

I’ve checked the videos and that’s unlikely the problem. Even if the hands dissappear the head is still there. And it would be very rare that in both videos I’ve tried the hands dissapear and appear from the frame 100 to 201, and as I said by visual check it seems that they are properly tracked during that period.

Luis.

6

Sorry for the multiple post, the prueba 10.mpg the grey areas are there on purpose because the tracker got confused often with the flag beneath it, so it’s not corrupted it’s that way :slight_smile:

7

Dear Luis, Przemek,
unfortunately I have no clue of what the problme is. In the next few days I’ll try reproducing the error (I am gonna download the video from dropbox) and see what causes his problem.
Przemek, do you have the same error if you run the recognizer on your laptop?
Best,
Stefano

8

Hi Stefano, Luis.

Indeed I managed to reproduce this error when I used the video and Skin Color values that you provided above. But when I load the video into the Skin Color Estimation App and choose my own values, the problem disappeared. Could you try to run the experiment with my file and report back if the problem disappeared?

If yes, we can try to figure out why exactly it appeared in the first place.

https://dl.dropboxusercontent.com/u/13823543/SKIN_prueba.xml

Best,
Przemek

9

With you file it worked fine. I’ve checked the files and the problem seems to be:

My File:

 <---- THIS LINE
	67
	164
	156
	128
	128
	18

Your file:

 <---- DIFFERENT TIMES
	135
	109
	158
	128
	15
	20
/span>

It seems like the Skin_FILE.xml does contain some information about the length of the video file, so it cannot be used with another video file. I was using the same Skin file for different videos as they are equal in colors but they are different in length.

And an extra problem, when it finishes tracking the hands, it gives some errors regarding the Tier files, and not being able to load them, this are quite random. I’m not concerned about it as I only need the coordinates, but just a Head Ups on it.

Cheers and thanks!

Luis.

10

Sorry I didn’t realize it would be interpreted :S

<!–

 &lt;--- Problem is here
	67
	164
	156
	128
	128
	18

–>

<!–

	135
	109
	158
	128
	15
	20

–>

11

My file:
/span start=“0” end=“6.24”/ <— Problem is here

Your file
/span start="0" end="17.928"/ <---- And here

Sorry for the multiple post (again) I didn't how to write it without being interpreted.

12

Thanks for the info Luis.
Indeed the skin xml file contains info about the start and endo fo the video. Based on the feedback from you and Przemek I will change the binaries so that the user will receive some info in case something like this happens.
Best,
Stefano