TRITRACK: Sparse Scene Flow Segmentation for Object Tracking
TriTrack (Triangulation based Generic Object Tracking from Stereo Sequences) has been developed by Philip Lenz at Karlsruhe Institute of Technology and is a simple and real-time capable algorithm that detects and tracks objects using scene flow as the only input. It does not depend on pre-trained category-specific object detectors. The library should run on Linux and Windows if you have MATLAB installed. See also: Sparse Scene Flow Segmentation for Moving Object Detection in Urban Environments (IV 2011).
Changelog
25.05.2014: Fixed some bugs and added some functionality to save raw detections in
KITTI format including a score for each detection
11.12.2013: First version online
Prerequisites
We tried to keep dependencies as small as possible, but to get started you are going to need at least:
MATLAB (requires the MATLAB statistics toolbox and uses the bgl library - binaries included)
libviso2 (for scene flow and egomotion computation)
TriTrack for Linux/Mac/Windows (including including a small test sequence)
Datasets
Sequences from the KITTI raw dataset can be used to test the algorithm
Citation
When using this software we will be happy if you cite the following related publications:
@inproceedings{Lenz2011IV,
author = {Philip Lenz and Julius Ziegler and Andreas Geiger and Martin Roser},
title = {Sparse Scene Flow Segmentation for Moving Object Detection in Urban Environments}, booktitle = {Intelligent Vehicles Symposium (IV)},
year = {2011}
}
@inproceedings{Geiger2011IV,
author = {Andreas Geiger and Julius Ziegler and Christoph Stiller},
title = {StereoScan: Dense 3D Reconstruction in Real-time}, booktitle = {Intelligent Vehicles Symposium (IV)},
year = {2011}
}