RGB-D Object Tracking

PEOPLE
  • Changhyun Choi
  • Henrik I. Christensen
  • ABSTRACT
    This paper presents a particle filtering approach for 6-DOF object pose tracking using an RGB-D camera. Our particle filter is massively parallelized in a modern GPU so that it exhibits real-time performance even with several thousand particles. Given an a priori 3D mesh model, the proposed approach renders the object model onto texture buffers in the GPU, and the rendered results are directly used by our parallelized likelihood evaluation. Both photometric (colors) and geometric (3D points and surface normals) features are employed to determine the likelihood of each particle with respect to a given RGB-D scene. Our approach is compared with a tracker in the PCL both quantitatively and qualitatively in synthetic and real RGB-D sequences, respectively.

    VIDEOS

    DATASET
    RGB-D Object Pose Tracking Dataset contains 4 synthetic and 2 real RGB-D image sequences which were used in the experiment of the paper "RGB-D Object Tracking: A Particle Filter Approach on GPU". Please cite the following paper if you use this dataset:
    @inproceedings{choi13iros_rgbdtracking,
      title = {{RGB}-D object tracking: {A} particle filter approach on {GPU}},
      booktitle = {Intelligent Robots and Systems ({IROS}), 2013 {IEEE}/{RSJ} International Conference on},
      author = {Choi, Changhyun and Christensen, Henrik I.},
      year = {2013},
      pages = {1084--1091},
    }
    Contents
    Name Description
    /README.txtA readme file explains the contents of this dataset
    /ground_truth.zip (0.2 MB)directory having ground truth motion trajectories for 4 synthetic sequences
    /models.zip (1.9 MB)directory containing mesh models (4 target objects and a kitchen models, all in PLY format)
    seq_synth_kinect_box_kitchen.7z (1.94 GB)synthetic RGB-D sequence of Kinect Box object
    seq_synth_milk_kitchen.7z (2.28 GB)synthetic RGB-D sequence of Milk object
    seq_synth_orange_juice_kitchen.7z (2.21 GB)synthetic RGB-D sequence of Orange Juice object
    seq_synth_tide_kitchen.7z (2.44 GB)synthetic RGB-D sequence of Tide object
    seq_real_milk_hand.7z (1.22 GB)real RGB-D sequence of Milk object
    seq_real_tide_hand.7z (1.32 GB)real RGB-D sequence of Tide object

    In "ground_truth" directory, there are "motion" files. Each motion file is a text file in which each line is corresponding to the pose of the object at time t=line number-1. Each line has 16 floating values which are from row-major SE(3) matrix. Please note that the pose represents position and orientation of the object with respect to the camera coordinate system.

    The "models" directory has a kitchen and 4 object models which were used in our work. They are PLY type, and thus they should be openned with viewers suppring PLY format (e.g. MeshLab).

    Sequences contain PCD (www.pointclouds.org) format files and are compressed with 7-zip which shows high compression ratio. You can find a free 7-Zip file archiver in http://www.7-zip.org/. Each sequence has 1000 frames, such as cloud0000.pcd, cloud0001.pcd, ..., cloud0999.pcd. The number in the file name means frame number.

    The default intrinsic parameters were used to set the projection matrix in OpenGL (GL_PROJECTION):
  • focal length: fx, fy = 525.0
  • principal point: ux = 319.5, uy = 239.5
  • lens distortion: none
  • PUBLICATIONS
    Changhyun Choi, Henrik I. Christensen, “RGB-D Object Tracking: A Particle Filter Approach on GPU,” in Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Tokyo Big Sight, Japan, 2013. [ pdf ]