ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question

Face Recognition

asked 2012-01-04 17:56:35 -0600

Peter Heim gravatar image

updated 2012-01-04 23:28:26 -0600

Hi All I just tried the new face recognition package but when i run Fserver i get the following error

[ INFO] [1325746855.901582077]: Could not load Haar cascade Face detection classifier in 'haarcascade_frontalface_alt.xml'.

I am using Ubuntu 10.4 32 bit and ros electric with vision opencv installed from debs

edit retag flag offensive close merge delete


Please post more information. There is not enough information here to reproduce your effrots. See
tfoote gravatar image tfoote  ( 2012-01-04 18:03:01 -0600 )edit

2 Answers

Sort by ยป oldest newest most voted

answered 2012-01-04 23:06:25 -0600

Pouyan gravatar image

updated 2012-01-04 23:15:02 -0600


Fserver requires haarcascade_frontalface_alt.xml to be in the same directory as you run the Fserver. The file was missing in the package, but now It has been uploaded in the repository in the face_recognition directory.

Cheers, Pouyan

edit flag offensive delete link more


Hi Just updated and did a rosmake the new files is in the face_recognition directory but i still get the same error
Peter Heim gravatar image Peter Heim  ( 2012-01-04 23:28:45 -0600 )edit
rosmake in not necessary again. you should just run the Fserver from the directory in which haarcascade_frontface_alt.xml is placed. For example : 1- copy haarcascade_frontface_alt.xml to /home/new_dir 2-cd /home/new_dir 3-rosrun face_recognition Fserver. Let me know if it works :)
Pouyan gravatar image Pouyan  ( 2012-01-04 23:37:35 -0600 )edit
You might also wish to put the data folder (containing sample pictures) and the train.txt file(which list the training examples) in the same directory. As all these are in the face_recognition directory by default, simply try : 1-roscd face_recognition 2-rosrun face_recognition Fserver
Pouyan gravatar image Pouyan  ( 2012-01-05 00:07:06 -0600 )edit
Running from the installation directory was the key the server started and i was able to acquire a face Thanks
Peter Heim gravatar image Peter Heim  ( 2012-01-05 00:07:53 -0600 )edit
Depending on the current directory being the package directory is very unfriendly for people using `roslaunch`. Perhaps a future release might provide a more flexible access mechanism?
joq gravatar image joq  ( 2012-01-05 02:05:58 -0600 )edit
As mentioned above, the program searches for its input files in its working directory which is not necessarily the package directory. In 'roslaunch', you can set the working directory using 'cwd' attribute to point to the directory in which necessary input files are placed.
Pouyan gravatar image Pouyan  ( 2012-01-05 08:11:39 -0600 )edit
That is really not the recommended way to use `roslaunch`. I am not even sure it works. Have you tried it?
joq gravatar image joq  ( 2012-01-05 09:14:54 -0600 )edit
I just tried it and i cant get it to work it shows a error when i use the directory if i use node or ROS_HOME and put the file in those locations i get the file not found error
Peter Heim gravatar image Peter Heim  ( 2012-01-05 09:33:41 -0600 )edit

answered 2012-03-14 01:22:11 -0600

nandhini gravatar image

hi pouyan, i'm using face_recognition package. i have changed threshold level between 70 to 80, because i felt default threshold is too heavy to detect a known person. threshold=88 works only when both person and camera are static. The problem is when an unknown person comes in the camera view, any name in the database is printing out like sergio, pouyan. can i get help to solve this issue.. Thank You.

edit flag offensive delete link more


I think you could use more training images for each face, lets say 100 instead of the current 25, to increase the precision and then playing with the threshold to get the best configuration.

Pouyan gravatar image Pouyan  ( 2012-03-16 01:37:22 -0600 )edit

which camera you used for training faces? just curious for checking camera clarity. i trained with 75 images initially, for a single person. trained persons are identified well and good. but problem persists with unknown person. The problem may be because of web camera i'm using and its clarity.

nandhini gravatar image nandhini  ( 2012-03-19 20:11:28 -0600 )edit

The face_recognition code is from Shervine Emami, you can find a link to his webpage in the face_recognition ROS wiki page. I haven't tested the performance extensively myself. Anyway, false positive answers for unknown persons should be eliminated using a higher confidence threshold.

Pouyan gravatar image Pouyan  ( 2012-03-21 04:02:55 -0600 )edit

Question Tools

1 follower


Asked: 2012-01-04 17:56:35 -0600

Seen: 1,465 times

Last updated: Mar 14 '12