VisOdomMonoPlaneInfinity (boofcv-sfm 0.33 API)

java.lang.Object
- boofcv.alg.sfm.d3.VisOdomMonoPlaneInfinity<T>

```
public class VisOdomMonoPlaneInfinity<T extends boofcv.struct.image.ImageBase<T>>
extends java.lang.Object
```
Estimates camera ego-motion by assuming the camera is viewing a flat plane and that off plane points are at infinity. These assumptions are often true for robots traveling over flat terrain outdoors. There are no assumptions for orientation of the camera, e.g. it could be upside down and at an angle relative to the plane.
A point feature tracker is used to extract sparse optical flow from the image sequence. Bad tracks are removed first through the use of geometric constraints and then through robust estimation techniques which search for maximum inlier sets. Before a track is discarded it has to be not used for X frames in a row, where X is user specified. Determination of inliers is done through a user specified threshold in pixels.
Points on the plane are used to estimate translation and rotation, while points at infinity are just used to estimate rotation. The two rotation estimates are merged together using a weighted vector average. Motion estimation does not require any off plane points at infinity, but does require points on the plane. All motion estimation is done internally in 2D, but can be converted into 3D.
It is possible to turn on a more rigid test for off plane points at infinity. If 'strictFar' is set to true then any points which contains motions that could not be generated by a rotation about the y-axis are discarded. This appears to be too strict and discard an excessive number of points when vehicle pitch up and down. Pitching motions will often not affect motion estimates from projected vectors on to the plane.

Nested Class Summary

Nested Classes
Modifier and Type Class and Description

static class VisOdomMonoPlaneInfinity.VoTrack
Additional track information for use in motion estimation

Nested Classes
Modifier and Type	Class and Description
`static class`	`VisOdomMonoPlaneInfinity.VoTrack` Additional track information for use in motion estimation

Constructor Summary

Constructors
Constructor and Description
`VisOdomMonoPlaneInfinity(int thresholdAdd, int thresholdRetire, double thresholdPixelError, org.ddogleg.fitting.modelset.ModelMatcher<georegression.struct.se.Se2_F64,PlanePtPixel> planeMotion, boofcv.abst.feature.tracker.PointTracker<T> tracker)` Configures motion estimation.

Method Summary

All Methods Static Methods Instance Methods Concrete Methods
Modifier and Type	Method and Description
`georegression.struct.se.Se2_F64`	`getCurrToWorld2D()` Returns the 2D motion estimate
`int`	`getTick()` Number of frames processed
`boofcv.abst.feature.tracker.PointTracker<T>`	`getTracker()`
`georegression.struct.se.Se3_F64`	`getWorldToCurr3D()` Converts 2D motion estimate into a 3D motion estimate
`protected boolean`	`isRotationFromAxisY(boofcv.abst.feature.tracker.PointTrack t, georegression.struct.point.Vector3D_F64 pointing)` Checks for motion which can't be caused by rotations along the y-axis alone.
`boolean`	`isStrictFar()`
`static double`	`maximizeCountInSpread(double[] data, int size, double maxSpread)` Finds the value which has the largest number of points above and below it within the specified spread
`boolean`	`process(T image)` Estimates the motion which the camera undergoes relative to the first frame processed.
`void`	`reset()` Resets the algorithm into its initial state
`void`	`setExtrinsic(georegression.struct.se.Se3_F64 planeToCamera)` Camera the camera's extrinsic parameters.
`void`	`setIntrinsic(boofcv.struct.calib.CameraPinholeBrown intrinsic)` Camera the camera's intrinsic parameters.
`void`	`setStrictFar(boolean strictFar)`

Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

- Constructor Detail
  - VisOdomMonoPlaneInfinity
```
public VisOdomMonoPlaneInfinity(int thresholdAdd,
                                int thresholdRetire,
                                double thresholdPixelError,
                                org.ddogleg.fitting.modelset.ModelMatcher<georegression.struct.se.Se2_F64,PlanePtPixel> planeMotion,
                                boofcv.abst.feature.tracker.PointTracker<T> tracker)
```
    Configures motion estimation.
    
    Parameters:
    
    thresholdAdd - New points are spawned when the number of on plane inliers drops below this value.
    
    thresholdRetire - Tracks are dropped when they are not contained in the inlier set for this many frames in a row. Try 2
    
    thresholdPixelError - Threshold used to determine inliers. Try 1.5
    
    planeMotion - Motion estimator for points on plane
    
    tracker - Image feature tracker
- Method Detail
  - setIntrinsic
```
public void setIntrinsic(boofcv.struct.calib.CameraPinholeBrown intrinsic)
```
    Camera the camera's intrinsic parameters. Can be called at any time.
    
    Parameters:
    
    intrinsic - Intrinsic camera parameters
  - setExtrinsic
```
public void setExtrinsic(georegression.struct.se.Se3_F64 planeToCamera)
```
    Camera the camera's extrinsic parameters. Can be called at any time.
    
    Parameters:
    
    planeToCamera - Transform from the plane to camera.
  - reset
```
public void reset()
```
    Resets the algorithm into its initial state
  - process
```
public boolean process(T image)
```
    Estimates the motion which the camera undergoes relative to the first frame processed.
    
    Parameters:
    
    image - Most recent camera image.
    
    Returns:
    
    true if motion was estimated or false if a fault occurred. Should reset after a fault.
  - isRotationFromAxisY
```
protected boolean isRotationFromAxisY(boofcv.abst.feature.tracker.PointTrack t,
                                      georegression.struct.point.Vector3D_F64 pointing)
```
    Checks for motion which can't be caused by rotations along the y-axis alone. This is done by adjusting the pointing vector in the plane reference frame such that it has the same y component as when the track was spawned and that the x-z components are normalized to one, to ensure consistent units. That pointing vector is then projected back into the image and a pixel difference computed.
    
    Parameters:
    
    pointing - Pointing vector of observation in plane reference frame
  - getTick
```
public int getTick()
```
    Number of frames processed
    
    Returns:
    
    Number of frames processed
  - getCurrToWorld2D
```
public georegression.struct.se.Se2_F64 getCurrToWorld2D()
```
    Returns the 2D motion estimate
    
    Returns:
    
    Motion estimate from current frame into world frame in 2D
  - getWorldToCurr3D
```
public georegression.struct.se.Se3_F64 getWorldToCurr3D()
```
    Converts 2D motion estimate into a 3D motion estimate
    
    Returns:
    
    from world to current frame.
  - maximizeCountInSpread
```
public static double maximizeCountInSpread(double[] data,
                                           int size,
                                           double maxSpread)
```
    Finds the value which has the largest number of points above and below it within the specified spread
    
    Parameters:
    
    data - Input data. Is modified by sort
    
    size - number of elements in data
    
    maxSpread - the spread it's going after
    
    Returns:
    
    best value
  - getTracker
```
public boofcv.abst.feature.tracker.PointTracker<T> getTracker()
```
  - isStrictFar
```
public boolean isStrictFar()
```
  - setStrictFar
```
public void setStrictFar(boolean strictFar)
```

Class VisOdomMonoPlaneInfinity<T extends boofcv.struct.image.ImageBase<T>>

Nested Class Summary

Constructor Summary

Method Summary

Methods inherited from class java.lang.Object

Constructor Detail

VisOdomMonoPlaneInfinity

Method Detail

setIntrinsic

setExtrinsic

reset

process

isRotationFromAxisY

getTick

getCurrToWorld2D

getWorldToCurr3D

maximizeCountInSpread

getTracker

isStrictFar

setStrictFar