filling the gaps of 3d mapping in monocular slam: from inverse depth...
TRANSCRIPT
Filling the gaps of 3D mapping in Monocular SLAM: from inverse depth for planes to super-pixels
Jose Martinez-Carranza
Associate Professor Royal Society-Newton Advanced Fellowship
Robotics Laboratory Computer Science Department
Instituto Nacional de Astrofisica Optica y Electronica (INAOE)
http://ccc.inaoep.mx/~carranza/
Twitter: @josemtzcarranza
Visual Simultaneous Localisation and Mapping
MonoSLAM Filtering Method based on EKF
PTAM Parallel Tracking and Mapping via
Optimisation
Simultaneous Mapping and Camera Pose Estimation
Separating Mapping from Camera Pose Estimation
Filling the Gaps in EKF SLAM Detecting planes in between map
points using visual appearance Incorporating planes within EKF
mapping (adaptive measurements)
• J. Martinez-Carranza, A. Calway. Efficiently Increasing Map Density in Visual SLAM Using Planar Features with Adaptive Measurements. Proceedings of the British Machine Vision Conference (BMVC). London, UK. September, 2009.
• J. Martinez-Carranza, A. Calway. Appearance Based Extraction of Planar Structure in Monocular SLAM. Proceedings of the Scandinavian Conference on Image Analysis (SCIA). Oslo, Norway. June, 2009.
Inverse Depth Plane Parameterisation
Follow inverse depth concept on initialising feature without delay and
wait for it to evolve
Inverse Depth for Feature Initialsiation in Monocular EKF SLAM
Inverse Depth Parameterisation for Planes
Inverse Depth Parameterisation for Planes
Inverse Depth Parameterisation for Planes
Inverse Depth Parameterisation for Planes
Inverse Depth Parameterisation for Planes
Inverse Depth Parameterisation for Planes
• J. Martinez-Carranza, A. Calway. Efficient Visual Odometry Using a Structure-Driven Temporal Map. Proceedings of the International Conference on Robotics and Automation (ICRA). Minnesota, USA. May, 2012.
• J. Martinez-Carranza, A. Calway. Unifying Planar and Point Mapping in Monocular SLAM. Proceedings of the British Machine Vision Conference (BMVC). Aberystwyth, UK. September, 2010.
Inverse Depth Parameterisation for Planes
IDPP and Plane Recognition
• O. Haines, J. Martinez-Carranza and A. Calway. Visual mapping using learned structural priors. Proceedings of the International Conference on Robotics and Automation (ICRA). Karlsruhe, Germany. May, 2013.
Plane Recognition/Orientation in a Single Image
Plane Recognition/Orientation in a Single Image
• J. A. Osuna-Coutiño, J. Martinez-Carranza, M. Arias-Estrada, W. Mayol-Cuevas. Dominant Plane Recognition in Interior Scenes from a Single Image. International Conference on Pattern Recognition. Cancun (ICPR), Mex. Dec, 2016.
Enhancing Mapping with Separate Processing
GPU Smart Camera GPU+Superpixels
Towards a Smart Camera
• R. de Lima, J. Martinez-Carranza, A. Morales-Reyes and R. Cumplido. Toward a smart camera for fast high-level structure extraction. Journal of Real-Time Image Proc (2017). Springer. https://doi.org/10.1007/s11554-017-0704-5
Towards a Smart Camera
Filling the Gaps with SuperPixels
• C. Cruz-Martinez, J. Martínez-Carranza, J. & W. Mayol-Cuevas. Real-time enhancement of sparse 3D maps using a parallel segmentation scheme based on superpixels. Journal of Real-Time Image Proc (2017). Springer. https://doi.org/10.1007/s11554-017-0707-2.
Filling the Gaps with SuperPixels
• Stochastic mapping enables efficient frame-to-frame processing, but specially when combined with high level structure detection/segmentation. – Let the hard work (of detection/segmentation/recognition) to be done by
machine learning approaches).
• Separating the mapping enables light computing to be carried out where suitable. – Camera pose estimates on low-budget computers/drones. – High cost processing on specialised hardware (GPU, FPGA).
Final Remarks