cirquit
PhD student with a focus on machine learning, distributed systems and functional programming.
Introduction to MonoSLAM
Introduction
1
Alexander Isenko
Part I
Goals
Building a map
2
(0,0)
(3.6, 2)
(?, ?)
Goals
Localization
3
(0,0)
(3.6, 2)
(5, 2.5)
(4, 5)
(3, 2.5)
SLAM Pipeline
Pipeline
4
Visual Sensors
Data
Association
Odometry
EKF
Odo. Update
EKF
Re-Observation
EKF New Observation
Landmark
Extraction
30 Hz
100 Hz
- image data
- edges
- features
known?
unknown?
- angle diff.
Landmarks
Landmarks
5
- should be unique
- re-observable
- from multiple angles
- stationary
- plentiful
Landmarks
Landmarks
6
Localization - MonoSLAM
Localization
7
Requirements:
- constant time calculation
- time-independent state calculation
- start with a known reference point
- work in a restricted volume
- any 3D translation should be possible
- loop closing in 10s or 10min, doesn't matter
- only map necessary features, we want localization
- stable, long - term landmarks
Structure from Motion
Localization
8
State
State Description
8
\hat{x}_p = \{ angle, position \}
\hat{y_i} = \{ angle, position \}
\hat{x}_p
\hat{y_1}
\hat{y_2}
\hat{x} = \{ x_p, y_1, y_2, ... , y_n \}
P_n = cov(\hat{x}, \hat{x})
State
State Description
9
\hat{x}_p
\hat{y_1}
\hat{y_2}
\hat{x} = \{ x_p, y_1, y_2, ... , y_n \}
P_n = cov(\hat{x}, \hat{x})
0.5
1.5
State
State Description
10
\hat{x}_p
\hat{y_1}
\hat{y_2}
\hat{x} = \{ x_p, y_1, y_2, ... , y_n \}
P_n = cov(\hat{x}, \hat{x})
P_n = \begin{bmatrix}
\hat{x}\hat{x} & \hat{x}y_1 & ... & \hat{x}y_n \\
y_1\hat{x} & y_1 y_2 & ... & y_1 y_n \\
... \\
y_n\hat{x} & y_n y_2 & ... & y_n y_n \\
\end{bmatrix}
0.8
2.0
Next Lession
State Description
11
Extended Kalman Filter
Covariance Visualized
Measurement Model
Prediction Model
How to encorparate noise
Introduction to MonoSLAM (Part I)
By cirquit
