Detects graph correspondences based on visual feature matching between keyframes. More...

#include <keyframe_graph_detector.h>

Public Member Functions
void	generateKeyframeAssociations (KeyframeVector &keyframes, KeyframeAssociationVector &associations)
	KeyframeGraphDetector (const ros::NodeHandle &nh, const ros::NodeHandle &nh_private)
	Constructor from ROS nodehandles.
virtual	~KeyframeGraphDetector ()
	Default destructor.
Protected Attributes
ros::NodeHandle	nh_
	the public nodehandle
ros::NodeHandle	nh_private_
	the private nodehandle
Private Member Functions
double	distEuclideanSq (const PointFeature &a, const PointFeature &b)
	Squared Euclidean distance between two features in 3D.
void	getRandomIndices (int k, int n, IntVector &output)
	Returns k distinct random numbers from 0 to (n-1)
void	manualBruteForceAssociations (KeyframeVector &keyframes, KeyframeAssociationVector &associations)
	Creates associations based on visual matching between keyframes through a RANSAC test (for manually added frames)
void	pairwiseMatchingRANSAC (RGBDFrame &frame_a, RGBDFrame &frame_b, double max_eucl_dist_sq, double max_desc_dist, double sufficient_inlier_ratio, std::vector< cv::DMatch > &all_matches, std::vector< cv::DMatch > &best_inlier_matches, Eigen::Matrix4f &best_transformation)
	Given two keyframes, finds all keypoint correposndences which follow a rigid transformation model.
void	prepareFeaturesForRANSAC (KeyframeVector &keyframes)
	Goes through all the keyframes and fills out the required information (features, distributinos, etc) which will be needed by RANSAC matching.
void	trainMatcher (const KeyframeVector &keyframes, cv::FlannBasedMatcher &matcher)
	Aggregates all features across all keyframes and trains a knn flann matcher in descriptor space.
void	treeAssociations (KeyframeVector &keyframes, KeyframeAssociationVector &associations)
	Creates associations based on visual matching between keyframes through a RANSAC test.
void	visualOdometryAssociations (KeyframeVector &keyframes, KeyframeAssociationVector &associations)
	Creates associations based on the visual odometry poses of the frames, ie, associations between consecutive frames only.
Private Attributes
int	k_nearest_neighbors_
	How many nearest neighbors are requested per keypoint.
double	max_corresp_dist_desc_
	Maximum distance (in descriptor space) between two features to be considered a correspondence candidate.
double	max_corresp_dist_eucl_
	Maximum distance (in Euclidean space) between two features to be considered a correspondence candidate.
double	max_corresp_dist_eucl_sq_
	Derived from max_corresp_dist_eucl_.
int	max_ransac_iterations_
	Maximim iterations for the RANSAC test.
int	min_ransac_inliers_
	How many inliers are required to pass the RANSAC test.
int	n_keypoints_
	Number of desired keypoints to detect in each image.
double	n_ransac_candidates_
	For kd-tree based correspondences, how many candidate keyframes will be tested agains the query keyframe using a RANSAC test.
std::string	ransac_results_path_
	The path where to save images if save_ransac_results_ is true.
bool	save_ransac_results_
	If true, positive RANSAC results will be saved to file as images with keypoint correspondences.

Detailed Description

Detects graph correspondences based on visual feature matching between keyframes.

Definition at line 41 of file keyframe_graph_detector.h.

Constructor & Destructor Documentation

ccny_rgbd::KeyframeGraphDetector::KeyframeGraphDetector	(	const ros::NodeHandle &	nh,
		const ros::NodeHandle &	nh_private
	)

Constructor from ROS nodehandles.

Parameters:

nh	the public nodehandle
nh_private	the private nodehandle

Definition at line 28 of file keyframe_graph_detector.cpp.

ccny_rgbd::KeyframeGraphDetector::~KeyframeGraphDetector ( ) [virtual]

Default destructor.

Definition at line 60 of file keyframe_graph_detector.cpp.

Member Function Documentation

double ccny_rgbd::KeyframeGraphDetector::distEuclideanSq	(	const PointFeature &	a,
		const PointFeature &	b
	)		`[inline, private]`

Squared Euclidean distance between two features in 3D.

Parameters:

a	the first 3D point
b	the second 3D point

Returns:: squared Euclidean distance between two a and b

Definition at line 172 of file keyframe_graph_detector.h.

void ccny_rgbd::KeyframeGraphDetector::generateKeyframeAssociations	(	KeyframeVector &	keyframes,
		KeyframeAssociationVector &	associations
	)

Main method for generating associatuions

Parameters:

keyframes	the input vector of RGBD keyframes
associations	reference to the output vector of associations

Definition at line 65 of file keyframe_graph_detector.cpp.

void ccny_rgbd::KeyframeGraphDetector::getRandomIndices	(	int	k,
		int	n,
		IntVector &	output
	)		`[private]`

Returns k distinct random numbers from 0 to (n-1)

Parameters:

k	the number of random samples
n	the (exclusive) upper limit of the number range
output	the output vector of random numbers

Definition at line 502 of file keyframe_graph_detector.cpp.

void ccny_rgbd::KeyframeGraphDetector::manualBruteForceAssociations	(	KeyframeVector &	keyframes,
		KeyframeAssociationVector &	associations
	)		`[private]`

Creates associations based on visual matching between keyframes through a RANSAC test (for manually added frames)

Performs a brute force (each-to-each) matching between all frames which have been manually added.

Parameters:

keyframes	the input vector of RGBD keyframes
associations	reference to the output vector of associations

Definition at line 148 of file keyframe_graph_detector.cpp.

void ccny_rgbd::KeyframeGraphDetector::pairwiseMatchingRANSAC	(	RGBDFrame &	frame_a,
		RGBDFrame &	frame_b,
		double	max_eucl_dist_sq,
		double	max_desc_dist,
		double	sufficient_inlier_ratio,
		std::vector< cv::DMatch > &	all_matches,
		std::vector< cv::DMatch > &	best_inlier_matches,
		Eigen::Matrix4f &	best_transformation
	)		`[private]`

Given two keyframes, finds all keypoint correposndences which follow a rigid transformation model.

Parameters:

frame_a	the first RGBD frame
frame_b	the second RGBD frame
max_eucl_dist_sq	maximum squared Euclidean distance (in meters) between two features in a correspondence for that correspondence to be considered an inlier in the transformation model
max_desc_dist	maximum descriptor distance between two features in a correspondence for that correspondence to be considered an inlier in the transformation model
sufficient_inlier_ratio	if the ratio between inlier matches to candidate matches exceeds this, the RANSAC test terminates successfuly
all_matches	output vector of all the matches
best_inlier_matches	output vectors of the matches which are inliers to the best transformation model
best_transformation	the best transformation determined by RANSAC

Definition at line 217 of file keyframe_graph_detector.cpp.

void ccny_rgbd::KeyframeGraphDetector::prepareFeaturesForRANSAC ( KeyframeVector & keyframes ) [private]

Goes through all the keyframes and fills out the required information (features, distributinos, etc) which will be needed by RANSAC matching.

Uses SURF features with an adaptive thershold to ensure a minimum number of feautres (n_keypoints_) detected in each frame

Parameters:

keyframes the vector of keyframes to be used for associations

Definition at line 78 of file keyframe_graph_detector.cpp.

void ccny_rgbd::KeyframeGraphDetector::trainMatcher	(	const KeyframeVector &	keyframes,
		cv::FlannBasedMatcher &	matcher
	)		`[private]`

Aggregates all features across all keyframes and trains a knn flann matcher in descriptor space.

Parameters:

keyframes	the input keyframes
matcher	the reference to the matcher

Definition at line 357 of file keyframe_graph_detector.cpp.

void ccny_rgbd::KeyframeGraphDetector::treeAssociations	(	KeyframeVector &	keyframes,
		KeyframeAssociationVector &	associations
	)		`[private]`

Creates associations based on visual matching between keyframes through a RANSAC test.

Candidates for RANSAC testing are determined by building a kd-tree of all the features (across all keyframe). Then for each keyframe, the tree is used to determine a number of possible candidate keyframes. RANSAC is performed between the query frame and the valid candidates.

Parameters:

keyframes	the input vector of RGBD keyframes
associations	reference to the output vector of associations

Definition at line 375 of file keyframe_graph_detector.cpp.

void ccny_rgbd::KeyframeGraphDetector::visualOdometryAssociations	(	KeyframeVector &	keyframes,
		KeyframeAssociationVector &	associations
	)		`[private]`

Creates associations based on the visual odometry poses of the frames, ie, associations between consecutive frames only.

Parameters:

keyframes	the input vector of RGBD keyframes
associations	reference to the output vector of associations

Definition at line 123 of file keyframe_graph_detector.cpp.

Member Data Documentation

int ccny_rgbd::KeyframeGraphDetector::k_nearest_neighbors_ [private]

How many nearest neighbors are requested per keypoint.

Definition at line 91 of file keyframe_graph_detector.h.

double ccny_rgbd::KeyframeGraphDetector::max_corresp_dist_desc_ [private]

Maximum distance (in descriptor space) between two features to be considered a correspondence candidate.

Definition at line 103 of file keyframe_graph_detector.h.

double ccny_rgbd::KeyframeGraphDetector::max_corresp_dist_eucl_ [private]

Maximum distance (in Euclidean space) between two features to be considered a correspondence candidate.

Definition at line 108 of file keyframe_graph_detector.h.

double ccny_rgbd::KeyframeGraphDetector::max_corresp_dist_eucl_sq_ [private]

Derived from max_corresp_dist_eucl_.

Definition at line 112 of file keyframe_graph_detector.h.

int ccny_rgbd::KeyframeGraphDetector::max_ransac_iterations_ [private]

Maximim iterations for the RANSAC test.

Definition at line 73 of file keyframe_graph_detector.h.

int ccny_rgbd::KeyframeGraphDetector::min_ransac_inliers_ [private]

How many inliers are required to pass the RANSAC test.

If a candidate keyframe has fewer correspondences or more, it will not be eligible for a RANSAC test

Definition at line 98 of file keyframe_graph_detector.h.

int ccny_rgbd::KeyframeGraphDetector::n_keypoints_ [private]

Number of desired keypoints to detect in each image.

Definition at line 116 of file keyframe_graph_detector.h.

double ccny_rgbd::KeyframeGraphDetector::n_ransac_candidates_ [private]

For kd-tree based correspondences, how many candidate keyframes will be tested agains the query keyframe using a RANSAC test.

Definition at line 87 of file keyframe_graph_detector.h.

ros::NodeHandle ccny_rgbd::KeyframeGraphDetector::nh_ [protected]

the public nodehandle

Definition at line 66 of file keyframe_graph_detector.h.

ros::NodeHandle ccny_rgbd::KeyframeGraphDetector::nh_private_ [protected]

the private nodehandle

Definition at line 67 of file keyframe_graph_detector.h.

std::string ccny_rgbd::KeyframeGraphDetector::ransac_results_path_ [private]

The path where to save images if save_ransac_results_ is true.

Definition at line 82 of file keyframe_graph_detector.h.

bool ccny_rgbd::KeyframeGraphDetector::save_ransac_results_ [private]

If true, positive RANSAC results will be saved to file as images with keypoint correspondences.

Definition at line 78 of file keyframe_graph_detector.h.

The documentation for this class was generated from the following files:

Public Member Functions

Protected Attributes

Private Member Functions

Private Attributes

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation