Visual Odometry

Visual Odometry

• Research Background

  • In the case of environments in which a GPS signal cannot be acquired (i.e forests, buildings, and concentrated urban areas), an alternative localization method is needed. 

  • Instead of using costly sensors like GPS or laser scanners, a camera is an inexpensive sensor with potential for various applications.

• Research Objectives

  • Using a stereo-camera, relative motion between each frame is estimated and robot trajectory is drawn.

  • Feature extraction, feature matching, outlier removal and motion estimation is the structure of this algorithm.

  • Furthermore, mono-camera and Kinect are to be utilized to reduce processing time and to deal with dark environments. 

  • Optimal algorithm that is robust in outdoor environments (where light levels and shadows must be accounted for) is also being researched.

  • In the future, drift-error will be resolved using landmarks. 

• Research Output

  • Improved accuracy was shown using visual odometry (red, stereo-camera) compared to odometry (blue, IMU+encoder)