mleotta / vehicle3d

Deformable 3D vehicle model for shape estimation and tracking from video (Ph.D. Thesis code for Matt Leotta)
Boost Software License 1.0
6 stars 2 forks source link

This software is the implementation accompanying the Ph.D. thesis of Matthew J. Leotta, "Generic, Deformable Models for 3-D Vehicle Surveillance", Brown University, May 2010.

For more information please visit the project website: http://www.lems.brown.edu/~mleotta/project/defvehicle3d.php This website has links to download the thesis and datasets used in the thesis.

Top level directory structure:

cmake CMake modules directory.

cmd Command line utilities. These are primarily used for training the deformable model. The python subdirectory provides simple scripts to run these utilities over batches of files.

data The learned data files. This includes PCA data files and a parts file.

dml Deformable Model Library. This library contains the all the functions and classes used in optimizing the vehicle shape from multiple views or video. The python subdirectory builds a module to provide a python interface to run experiments.

gui Graphical User Interface. This directory contains wxWidgets based GUI for running experiments and visual debugging.

klt Kanade-Lucas-Tomasi Feature Tracker. This is 3rd party code written by Stan Birchfield. It has its own license, see the klt/README file.

spl Stream Processing Library. This library provides a video stream processing framework.