CEC Pose

Pose detection made easy!

This is a plug-and-play package for pose detection. Originally developed for the TMM-HRI project, it extracts the pose of a person given an RGB image, a depth image, the person's bounding box and segmentation mask, and the robot/camera's own pose.

The library was designed to be as easy to use as possible.

Install

This package uses setuptools to make it accessible across your system.

Open a terminal and clone the git repository anywhere in your system, ideally not in the same folder as your project code.

git clone https://github.com/gt-cec/cec-pose

Install the package.

pip install -e cec-pose

And that's it!

Usage

Using the package is quite easy, here is an example:

# import the package, this will cause some spam from mmengine
import cec_pose

# create a pose detector object, this will cause some spam from mmengine
pose_detector = cec_pose.PoseDetector()

# obtain a person's pose in 3D space
pred_person_loc, predicted_heading, (rw_coordinates, mean_depth, first_person_3d, keypoints, pred_skeleton) = pose_detector.get_heading_of_person(rgb, depth, detected_humans, robot_pose)

The outputs of get_heading_of_person are:

• pred_person_loc
• predicted_heading
• rw_coordinates
• mean depth