Inverse Depth EKF MonoSLAM with optical flow

A monoSLAM program written in C++. This algorithm uses an Extended Kalman Filter (EKF) for camera state and landmark position estimation. The only sensor is a single camera using OpenCV's optical flow to track the landmarks. I am manually picking the points, but it is also set up to choose strong corners automatically when the landmarks drop below a given number. The ellipsoids represent an uncertainty of 2 sigma, or approximately the 95% confidence interval. The actual update rate is about 30 Hz by limit of the webcam, even though the recorded video rate is much slower. I eventually plan to use this as a downward looking camera for tracking ground features to improve UAV navigation and orientation solutions. The vision data will be integrated with inertial and GPS data to improve accuracy.

Category:

People & Blogs

Tags:

• MonoSLAM
• Inverse Depth
• Optical Flow
• SLAM
• Extended Kalman Filter
• Computer Vision
• Covariance
• Covariance Ellipsoid
• EKF
• Estimation