ros_openpose_gossip

1. Description

This package gives information on people based on openpose messages

openpose_gossip gives informations on people sorted from left to right such as :

- id                # num of the person in the image (from left to right)
- posture           # Possible values: "Standing" "Sitting" "Lying" "Undefined"
- handPosture       # Values : "Undefined" "Call" "Pointing Left" "Crossed" "Pointing Right" 
- boundingBox       # coordinates of a person bounding box
- shirtRect         # coordinates of shirt rectangle sample 
- trouserRect       # coordinates of trouser rectangle sample
- distanceEval      # Distance evaluation in meters from 2D informations (see 5.  Calibration)
- pose              # Pose of the person's neck, on robot local map
- faceConfidence    # Face confidence
- orientation       # person orientation : "Front" "Back" "Right" "Left" "Front Right" "Front Left"

This program was done for the pepper robot (front camera) with some specifications :

• /!\ posture needs camera to be parallel to the ground
• /!\ distanceEval needs a calibration from csv files (see 5. Calibration)

2. Authors

• Raphaël LEBER

3. Dependencies

openpose
ros_openpose (fork)

4. Quick start

4.1 Launch the openpose service

   roslaunch openpose_ros_node serviceReadyTest.launch

4.2 Run the Gossip service

   rosrun ros_openpose_gossip_node OpenPoseGossip_node.py

4.3 You can try the Gossip service with one of the following test or do your own call :

   roslaunch ros_openpose_gossip_examples test_on_pepper.launch
   roslaunch ros_openpose_gossip_examples test_on_webcam.launch
   roslaunch ros_openpose_gossip_examples test_with_file.launch

5. Calibration

If you use this package with another camera than the front camera of the pepper, you will need to do a calibration. It consists of recording the limbs length of an "average" person at a known distance, in a standing position, maximizing (with a natural front pose) each limbs length. Function SaveLimbsProfil can be used but is not implemented in a calibration program (TODO).
The algorithm will then compare the longest limb (which means with the highest 2D components) with the limb in one of the CSV file (explanation simplified).