Class FeatureTracker

Class Documentation

class FeatureTracker

Visual front end that processes LK optical flow tracking for each feature of each camera.

Public Functions

FeatureTracker()
map<int, vector<pair<int, Eigen::Matrix<double, 7, 1>>>> trackImage(double _cur_time, const cv::Mat &_img, const cv::Mat &_img1 = cv::Mat(), const cv::Mat &_mask = cv::Mat())

Generate feature properties based on input images.

Parameters

  • _cur_time – timestamp of the input image.

  • _img – Input rectified left image (cv::Mat)

  • _img1 – Input rectified right image (cv::Mat)

  • _mask – (Optional) Input mask (to filter out dynamic objects)

Returns

Special data structure to store features.

void setMask()

1) Set small masks around last tracked features, with a predefined radius 2) Set “cubicle” mask for dynamic objects.

void readIntrinsicParameter(const vector<string> &calib_file)

Read intrinsic parameters from config file.

Parameters

calib_file – config file location

void showUndistortion(const string &name)
void rejectWithF()

Fundamental Matrix Estimation with RANSAC to filter out outliers.

vector<cv::Point2f> ptsVelocity(vector<int> &ids, vector<cv::Point2f> &pts, map<int, cv::Point2f> &cur_id_pts, map<int, cv::Point2f> &prev_id_pts)

Calculates the 2D undistorted point velocity.

Parameters

  • ids – [in] Input vector of IDs

  • pts – [in] Input vector of 2D point coordinates

  • cur_id_pts – [out] std::map of ID and current 2D point coordinates

  • prev_id_pts – [in] std::map of ID and previous 2D point coordinates

Returns

Calculated vector of point velocities of each point

void showTwoImage(const cv::Mat &img1, const cv::Mat &img2, vector<cv::Point2f> pts1, vector<cv::Point2f> pts2)
void drawTrack(const cv::Mat &imLeft, const cv::Mat &imRight, vector<int> &curLeftIds, vector<cv::Point2f> &curLeftPts, vector<cv::Point2f> &curRightPts, map<int, cv::Point2f> &prevLeftPtsMap)

Display visualization of image overlaid by optical flows of tracked features.

Parameters

  • imLeft –

  • imRight –

  • curLeftIds – Vector of current left feature IDs.

  • curLeftPts – Vector of current left 2D feature point coordinates.

  • curRightPts – Vector of current right 2D feature point coordinates

  • prevLeftPtsMap – Vector of previous left 2D point coordinates

void setPrediction(map<int, Eigen::Vector3d> &predictPts)

Predict 2D points based on the 3D predicted points projection.

Parameters

predictPts – [in] std::map of predicted 3D points associated with IDs @Note There is no return value, but the predict_pts value is the hidden variable to be modified.

void removeOutliers(set<int> &removePtsIds)

Removes the invalid points of vector of “prev_pts”, “ids” and “track_cnt” based on flags.

Parameters

removePtsIds – the set of flags whether the points are outliers or not.

cv::Mat getTrackImage()

Inquire the tracked image from external files.

Returns

TrackImage after drawTrack() is implemented.

bool inBorder(const cv::Point2f &pt)

Judge whether a 2D pixel coordinate cv::Point2f is inside the image.

Parameters

pt – The pixel to be judged.

Returns

flag true if it is inside the image; false otherwise.

Public Members

int row
int col
cv::Mat imTrack
cv::Mat mask
cv::Mat prev_img
cv::Mat cur_img
cv::Mat dy_mask
cv::Mat dilate_mask_inv
vector<cv::Point2f> n_pts
vector<cv::Point2f> predict_pts
vector<cv::Point2f> predict_pts_debug
vector<cv::Point2f> prev_pts
vector<cv::Point2f> cur_pts
vector<cv::Point2f> cur_right_pts
vector<cv::Point2f> prev_un_pts
vector<cv::Point2f> cur_un_pts
vector<cv::Point2f> cur_un_right_pts
vector<cv::Point2f> pts_velocity
vector<cv::Point2f> right_pts_velocity
vector<int> ids
vector<int> ids_right
vector<int> track_cnt
map<int, cv::Point2f> cur_un_pts_map
map<int, cv::Point2f> prev_un_pts_map
map<int, cv::Point2f> cur_un_right_pts_map
map<int, cv::Point2f> prev_un_right_pts_map
map<int, cv::Point2f> prevLeftPtsMap
vector<camodocal::CameraPtr> m_camera
double cur_time
double prev_time
bool stereo_cam
int n_id
bool hasPrediction

Public Static Functions

static vector<cv::Point2f> undistortedPts(vector<cv::Point2f> &pts, camodocal::CameraPtr cam)

Undistort 2D points.

Parameters

  • pts – vector of 2D image points

  • cam – camera model of camodocal

Returns

The vector of undistorted points