awesomebytes/pi_vision

Pi Vision ported to ROS Indigo

pi_vision

Pi Vision ported to ROS Indigo, together with various enhancements:

• Its been catkinized.
• Can track multiple faces.
• Attempts to perform localization in 3D, so that face positons are now published in 3D coordinates.

This is based on the original Pi Vision package, taken from http://wiki.ros.org/pi_vision, which had been abandonded after groovy. This package includes both includes http://wiki.ros.org/pi_face_tracker and http://wiki.ros.org/pi_face_tracker_gui

The plain, unenhanced port of pi_vision to indigo can be found in the branch pi-vision-orig-indigo.

Installation

Requirements:

• ros-indigo-openni-camera
• mjpeg_server
• usb_cam

apt-get install ros-indigo-cv-bridge ros-indigo-image-transport
apt-get install ros-indigo-mjpeg-server
apt-get install ros-indigo-openni-camera
apt-get install ros-indigo-usb-cam

The node must be built from git:

git clone https://github.com/hansonrobotics/pi_vision
cd catkin; catkin build
source devel/setup.bash

Run

To change camera settings modify the usb_cam.launch file.

roslaunch ros2opencv usb_cam.launch 
roslaunch pi_face_tracker face_tracker_usb_cam.launch

ROS Topics

Faces /face_locations

Publishes a list of human faces being tracked. Each is given an ID number, and a 3D coordinate. The coordinate frame used is the usual ROS 'engineering' frame: x is straight ahead, y is the the left, and z is up. Units are in meters.

FaceEvent /face_event

Publishes face tracking events. Currently, the only events published are new_face and lost_face. The first indicates a newly-acquired face to track, the second, that a face is no longer visible.

TODO

1. Make it work with Kinect.
2. Migrate to Python cv2 wrappers from cv. The cv wrappers use a c++-like interface, the cv2 wrappers use numpy and are more efficient.