Skip to content
dzoberti edited this page Jan 8, 2016 · 12 revisions
  • Alpha 3.0: 2016-01-08

    • Fixed bug in reprojection to images.
    • Exiting iterations in ICP to prevent unnecessary iterations.
    • Added flag in config file for floor learning.
    • Improved interface.
    • Fixed bug in reprojection to images.
    • Store some information in log.
    • Fixed training and set generation in trainingApp.
    • Using EKF position and compass orientation for ICP guess. Before we used the previous ICP result or EKF position if ICP failed.
    • Floor angle calculation fixed and parameter added to config file.
  • Alpha 2.3: 2016-01-02

    • Added floor substraction learned online from first images.
    • Improved ICP adding extra iteration.
    • Improved template matching adding maxScore threshold
    • Added multiscale training in BoW
    • Cleaned multiple references of Coordinate system
    • Added EKF to integrate IMU data from Drone IMU and Stereo Vision Odometry
    • Cleaned mainApplication pipeline
    • Improved dataset recorder
    • Removed members in stereo cameras related to camera position
    • Improved map generation pipeline checking quality of results in ICP and using also EKF results.
  • Alpha 2.2: 2015-12-08

    • Moved common files to an external library.
    • Tested Ekf that joins imu and "faked icp" position.
    • Tested recordDataset application with real UAV.
  • Alpha 2.1: 2015-12-01

    • Forked cjson to be able to solve small fixes. Fixed parsing errors in cjson.
    • Created application to record images and imu information in one dataset.
  • Alpha 2.0: 2015-11-25

    • Fixed drawn of bound boxes of candidates.
    • Fixed crop of candidates when they are out of camera range.
    • Keeping candidates over time.
    • Comparing Candidate position with ground truth.
    • Added HoG as possible feature descriptor.
    • Using libsvmpp library as SVM tool for machine learning and object categorization.
    • Creation of histogram of features using OpenCV.
    • Improved blur detection method and fixed thresholds.
    • Improved localization of the camera.
    • Improved GUI (3d, plot of time, etc.)
Clone this wiki locally