Search Results

Search Constraints

Start Over
Filtering by: Keyword Robotics Remove constraint Keyword: Robotics Resource type Dataset Remove constraint Resource type: Dataset

Search Results

  • FCAV Simulation Dataset

    work
    Creator:
    Vasudevan, Ram, Barto, Charles, Rosaen, Karl, Mehta, Rounak, Matthew, Johnson-Roberson, and Nittur Sridhar, Sharath
    Description:
    A dataset for computer vision training obtained from long running computer simulations
    Keyword:
    autonomous driving, simulation, Computer Vision and Pattern Recognition, deep learning, Computer Science, object detection, and Robotics
    Discipline:
    Engineering

  • Michigan Indoor Corridor Dataset

    work
    Creator:
    Tsai, Grace and Kuipers, Benjamin
    Description:
    ******Michigan Indoor Corridor 2012 Dataset****** This dataset is made available for research purpose only. Please contact Grace Tsai( gstsai@umich.edu) for any questions or comments. This dataset was used to produce the results in our IROS 2012 paper. If you use the data, please cite the following reference in your publications related to this work: Grace Tsai and Benjamin Kuipers Dynamic Visual Understanding of the Local Environment for an Indoor Navigating Robot International Conference on Intelligent Robots and Systems (IROS'12) October 2012 The dataset contains 4 video sequences acquired with camera mounted on a wheeled vehicle. The camera was set-up so that there was zero tilt and roll angle with respect to the ground. The camera has a fixed height (0.47 m) with the ground throughout the video. The intrinsic parameters of the cameras are: Focal length fc = [ 1389.182714 1394.598277 ] Principal point cc = [ 672.605430 387.235803 ] The distortion of the camera has been corrected. For each video sequences, an estimated camera pose in each frame of the video is provided in the file pose.txt. Each line in the file looks like: <frame index> <x (pose)> <y (pose)> <theta (pose)> Note the camera poses provided here are estimated by using an occupancy grid mapping algorithm with a laser range finder to obtain the robot pose. The dataset provides a ground truth labeling for all the pixels every 10 frames for each video. The labels of each frame is stored as a 2D matrix in a .mat file. The filename of each .mat file corresponds to the image frame. The labels can be interpreted as followed: -2 -> ceiling plane -1 -> ground plane >0 -> walls The labels of the walls are illustrated in a .pdf figure. Note the figure is only a illustration graph, not an actual floor plan.
    Keyword:
    Robotics and Computer vision
    Discipline:
    Engineering