1 / 22

MonoSLAM: Real-Time Single Camera SLAM

marty
Download Presentation

MonoSLAM: Real-Time Single Camera SLAM

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


    1. Pedro Davalos MonoSLAM 1 MonoSLAM: Real-Time Single Camera SLAM Davison, Reid, Molton, Stasse – IEEE, 2007 Presented By: Pedro Davalos February 25, 2008

    2. Pedro Davalos MonoSLAM 2 Localization: Where am I?

    3. Pedro Davalos MonoSLAM 3 Mapping: Symbolic Representation of the World

    4. Pedro Davalos MonoSLAM 4 Problem Position Estimation Localization/Mapping Dilemma: How to determine my Location? Use a Map! How do I build a Map? Knowing my Location!

    5. Pedro Davalos MonoSLAM 5 Mapping Examples

    6. Pedro Davalos MonoSLAM 6 Where Am I?

    7. Pedro Davalos MonoSLAM 7 Background: SLAM Simultaneous Localization And Mapping “SLAM is concerned with the problem of: -building a map of an unknown environment by a mobile robot while at the same time -navigating the environment using the map.” Early SLAM: A Stochastic Map for Uncertain Spatial Relationships [Smith, Self, Cheeseman 1987] Directed Sonar Sensing for Mobile Robot Navigation [Leonard 1990] (PhD) An Information-Theoretic Approach to Data Fusion and Sensor Management [Manyika 1993] SLAM [Csorba 1997] (PhD) Mobile Robot Navigation Using Active Vision [Davison 1998] (PhD)

    8. Pedro Davalos MonoSLAM 8 Background: SLAM Landmark Extraction Data Association State Estimation State Update & Landmark Update

    9. Pedro Davalos MonoSLAM 9 SLAM Demo

    10. Pedro Davalos MonoSLAM 10 Approach Implement Solution for Localization Use SLAM Technique With a single Camera (320x240) – calibrated, 100° FOV Real-Time (30Hz) Repeatable (drift free)

    11. Pedro Davalos MonoSLAM 11 MonoSLAM: Probabilistic 3D Map

    12. Pedro Davalos MonoSLAM 12 MonoSLAM: Natural Visual Landmarks Find/add new landmarks (Image Processing) Scan full image to find salient features [Shi, Tomasi, 1994] Save Landmarks Save 11x11 template AND position & orientation*

    13. Pedro Davalos MonoSLAM 13 MonoSLAM: System Initialization 3D from single frame? Startup with know prior reference target in the scene Provides scale and depth Allows for immediate normal operation mode

    14. Pedro Davalos MonoSLAM 14 MonoSLAM: Motion Model and Prediction Agile Camera with unknown dynamics Assume: constant velocity and constant angular velocity Impose smoothness in motion (assume unlikely large accelerations)

    15. Pedro Davalos MonoSLAM 15 Active Feature Measurement and Map Update Search and find features by cross-correlation Minimize search by predicting location on image Uncertainty in prediction of feature location

    16. Pedro Davalos MonoSLAM 16 MonoSLAM: Feature Initialization Identify a distinctive feature Save camera position, ray from camera to feature, and template As the camera moves, triangulate position if feature is reobserved

    17. Pedro Davalos MonoSLAM 17 MonoSLAM: Map Management Decide when to search for new features? If less than 12 features in the FOV Add one feature by searching 80x60 region without features, centric Decide when to delete bad features If 50% failure of recognizing features that should be visible Due to occlusion, moved, specularity,

    18. Pedro Davalos MonoSLAM 18 MonoSLAM: Feature Orientation Estimation Goal: Optimize template matching invariant to scale, rotation, translation Approach: Assume template is plane Based on Cam position Predict template normal Update surface n estimate Warp template with Homography

    19. Pedro Davalos MonoSLAM 19 Test Results Augmented Reality RealTime tracking 3d position allows 3d overlay

    20. Pedro Davalos MonoSLAM 20 Results Humanoid Used remote controlled Humanoid robot Tracked Position in realtime, 30Hz using: Single Camera Also used gyro!

    21. Pedro Davalos MonoSLAM 21 Limitations Blank Walls Lost Robot 100 Features Small area: room Heuristics: Thresholds Estimate Uncertainty

    22. Pedro Davalos MonoSLAM 22 Discussion?

More Related