handtrack-evaluation

Evaluation system for handtracking from RGB, in comparison with Leap Motion

Leap Motion

Dependencies

1. To get the Leap Motion set up, first sign up for a developer account on https://www.leapmotion.com.
2. Once you have an account, log in and navigate to the SDK tab and download the V2 SDK for Mac OS and Linux.

Project Installation

Change the following lines in leap_motion/leap_evaluation.py

1. On Line 4, set TEXT_FILENAME to whatever you want to call the text file. The default is leap_position.txt. Unless you set it, it will create a textfile in the same directory as leap_evaluation.py.
2. On Line 5, set PATH_TO_LEAP_SDK to the absolute path to the Leap Motion SDK lib directory.
3. On Lines 5-6, set the window_width and window_height to the dimensions of the video you plan to pass into the PerPixel handtracker.

If you have a python distribution from MacPorts or Homebrew you will need to use the install name tool to point to the correct version of python. Follow the steps in this [tutorial] (https://developer.leapmotion.com/documentation/python/devguide/Project_Setup.html?proglang=python).

Evaluation

Steps to run the evaluation script

1. The script can be found at evaluation/compare.py
2. Run python compare.py <path to ground truth> <path to algorithm> in the shell to compare the performace using the fingertip positions from two algorithms
3. The script outputs a mean-squared error (MSE) averaged across the number of frames used

Ground Truth Labeler

Used to obtain ground truth (x,y) coordinates for the fingertip positions.

1. The script can be found at label_tool/labeler.py
2. Run python labeler.py --input <InputDir> --output <OutputDir>, where the input directory contains the images to be labeled
3. Make sure to label the fingertips in the same order, as extracted using any tracking algorithms. Currently, the tracking algorithm finds the fingertips with the highest y-coordinate to the lowest
4. The script contains additional information for the labeler tool