Robust pose estimation for walking robot

Document Sample
Robust pose estimation for walking robot Powered By Docstoc
					Robust pose estimation
     for walking robot

                  M.Sc. Mauricio Edgar Stivanello
 Dr.ir. M.J.G. van de Molengraft (René) - Advisor
               Agenda
               Robust pose estimation for walking robot




●   Introduction
●   Camera model and Calibration
●   Pose Estimation
●   Future steps and Final considerations




                                                          2
                        Introduction
                        Robust pose estimation for walking robot




●   Walking robot
     –   Legs instead of wheels
     –   Pose estimation with 6-DOF
●   Goal
     –   Calibration based on vision
●   Challenges
     –   Direct image-world distance mapping is not
         available
     –   Orthographic projection of the ground plane is
         not possible
           ●   Image Remapping is not possible

                                                                   3
               Agenda
               Robust pose estimation for walking robot




●   Introduction
●   Camera model and Calibration
●   Pose Estimation
●   Future steps and Final considerations




                                                          4
                        Camera Model and Calibration
                        Robust pose estimation for walking robot



●   Description of the camera and
    mirror
     –   Relation between image and
         world points
     –   Unified Model
          ●   Set of Parameters
          ●   Camera and mirror abstraction
          ●   Suitable to different systems




                                                                   5
                       Camera Model and Calibration
                       Robust pose estimation for walking robot




●   Camera and Mirror Calibration
    –   OcamCalib Toolbox
         ●   Compatible with Unified Model
         ●   Simple planar pattern
         ●   Simple calibration procedure




                                                                  6
                       Camera Model and Calibration
                       Robust pose estimation for walking robot




●   Camera and Mirror Calibration
    –   OcamCalib Toolbox
         ●   Compatible with Unified Model
         ●   Simple planar pattern
         ●   Simple calibration procedure




    –   Image on sphere




                                                                  7
               Agenda
               Robust pose estimation for walking robot




●   Introduction
●   Camera model and Calibration
●   Pose Estimation
●   Future steps and Final considerations




                                                          8
                      Pose Estimation
                      Robust pose estimation for walking robot




●   Direct Linear Transformation
     –   Based on image-world
         correspondences
     –   Basic equation

                          m≃[ R t ] M w
     –   For n markers

                            Bp=0




                                                                 9
                         Pose Estimation
                         Robust pose estimation for walking robot




●   Direct Linear Transformation

     –   Minimization using SVD                         min p∥Bp∥2

     –   Sensible to errors
           ●   Normalization


     –   Practical problem
           ●   Image-World correspondences are needed




                                                                     10
                        Pose Estimation
                        Robust pose estimation for walking robot




●   Direct Linear
    Transformation
    –   Assumption:
          ●   Markers available
    –   Manual initialization
    –   Point tracking
          ●   Lucas and Kanade




                                                                   11
                      Pose Estimation
                      Robust pose estimation for walking robot




●
    Direct Linear
    Transformation
    –   Translation

         E max =5 %




                                                                 12
                   Pose Estimation
                   Robust pose estimation for walking robot




●
    Direct Linear
    Transformation
    –   Rotation




                                                              13
                         Pose Estimation
                         Robust pose estimation for walking robot




●   Egomotion
    –   Instantaneous rotation and translation
    –   Based on frame-to-frame changes
    –   2 main steps:
          ●   Calculation of motion fields: Motion field
              using original images (Optical flow)
          ●   Motion Estimation: Different algorithms
              (Bruss and Horn)


    –   Practical problem
          ●   Integration introduces additive error
              at each step


                                                                    14
                        Pose Estimation
                        Robust pose estimation for walking robot




●   Egomotion
    –   Motion fields on the
        soccer field


    –   Observed Problems
          ●   Reduced number of reliable
              points
          ●   Computational cost




                                                                   15
               Agenda
               Robust pose estimation for walking robot




●   Introduction
●   Camera model and Calibration
●   Pose Estimation
●   Future steps and Final considerations




                                                          16
                         Next steps and Final Comments
                         Robust pose estimation for walking robot




●   DLT
    –   Refinement of results
          ●   Nonlinear optimization
          ●   Minimize of the re-projection error
    –   Robust Markers identification


●   Egomotion
    –   Isolation of reliable points
    –   Reduce the time consumption




                                                                    17
                   Thank you!

              M.E.Stivanello@tue.nl
              maustiva@das.ufsc.br




     Dynamics and Control Technology group
     TU/e - Technische Universiteit Eindhoven
Department of Automation and Systems Engineering
   UFSC – Federal University of Santa Catarina