TO DO list

TO DO!

For now:

• 3D point -> remove camera distortion every time (especially before depth usage)
• Prepare program to use 2D features
• Do the evaluation instruction
• Kinect/Xtion/Pgrey -> initPose from file (right now everytime identity)
• LDB as 3rd party
• Detector parameters
• Core dumped when closing in matching
• Detection on BGR image
• DBScan
• removing features with depth > 6m
• test solution using mean error in RANSAC
• depthless features in optimization
• octomap from ROS
• Use some kind o adjusting mechanism in detection step
• Separate folder for Demo sources
• Uncertainty model in RANSAC
• RANSAC on reprojection error
• Garbage collector in map
• Cechy pobierane jedynie z póz bliskich euklidesowo/kątowo
• Usage of patches; incremental correction based on patches
• What happens when VO is lost
• Main PUTSLAM class
• Remove PCL dependency

Sprint till May:

• new ORB implementation based on ORBSLAM
• Umeyama vs g2o
• Motion model
• Point of view threshold -> more than 17,2 deg
• features from map (What does two thresholds in ORBSLAM do?)
• project features on an image that is a little bit bigger
• garbage collector
• keyframe -> removing unnecessary measurements -> so called policemen :)
• Draw map.size() vs frame no. (M.Nowicki)
• Delay VO to have optimized graph (M.Nowicki)
• getVisibleFeatures - smarter!
• Loop Closure (M.Nowicki)
• backend cleaning (M.Nowicki)
• MiT stata (M.Nowicki)
• optimization on reprojection error
• separate parameters for Tracking and LC

Later:

• OpenCV
• ROS
• Dataset MiT
• Visualization OpenGL

KLT parameters used in SVO:

• winSize = 30
• maxIter = 30
• eps = 0.001
• maxLvl = 4