face_pose_estimation.py

What is this?

https://user-images.githubusercontent.com/1901008/31308440-f7ffb6a6-abb1-11e7-8b89-0a332001448b.png

Estimate people face pose / gender from RGB Image and PeoplePoseArray. Please refer to Hyperface.

Subscribing Topic

  • ~input (sensor_msgs/Image)

    Input image.

  • ~input/pose_2d (jsk_recognition_msgs/PeoplePoseArray)

    Input people pose in 2D input image.

  • ~input/pose (jsk_recognition_msgs/PeoplePoseArray)

    Input people pose in 3D.

Publishing Topic

  • ~output/pose (geometry_msgs/PoseArray)

    Estimated face poses for each people.

  • ~output/gender (jsk_recognition_msgs/ClassificationResult)

    Estimated gender with confidence for each people.

  • ~output/rects (jsk_recognition_msgs/RectArray)

    Face rectangles in 2D input image used for cropping.

These 3 topics are need to be synchronized.

Parameters

  • ~gpu (Int, Default: -1)

    GPU id. -1 represents CPU mode.

  • ~classifier_name (String, Default: rospy.get_name())

    Name of this classifier

  • ~approximate_sync (Bool, Default: False)

    Use approximate synchronization policy.

  • ~queue_size (Int, Default: 100)

    Queue size for synchronization.

  • ~slop (Double, Default: 0.1)

    Slop for approximate sync.

  • ~face_padding (Double, Default: 0.0)

    Padding size factor for face rectangles.

  • ~face_threshold (Double, Default: 0.5)

    Threshold for confidence of detected faces.

Example

roslaunch jsk_perception sample_face_pose_estimation.launch GPU:=0