ORBextractor

​

Overview

​

Constructor

​

ORBextractor

ORBextractor(int nfeatures, float scaleFactor, int nlevels,
             int iniThFAST, int minThFAST)

• nfeatures - Target number of features to extract
• scaleFactor - Scale factor between pyramid levels (typically 1.2)
• nlevels - Number of pyramid levels
• iniThFAST - Initial FAST detection threshold
• minThFAST - Minimum FAST threshold if initial detection fails

​

Feature Extraction

​

operator()

int operator()(cv::InputArray _image, cv::InputArray _mask,
               std::vector<cv::KeyPoint>& _keypoints,
               cv::OutputArray _descriptors,
               std::vector<int> &vLappingArea)

• _image - Input image
• _mask - Mask for extraction region (currently ignored)
• _keypoints - Output vector of detected keypoints
• _descriptors - Output matrix of ORB descriptors
• vLappingArea - Lapping area information for stereo

• Features are dispersed uniformly using an octree structure
• Extracts features across multiple scale levels
• Computes 256-bit ORB binary descriptors

​

Scale Information

​

GetLevels

int GetLevels()

​

GetScaleFactor

float GetScaleFactor()

​

GetScaleFactors

std::vector<float> GetScaleFactors()

​

GetInverseScaleFactors

std::vector<float> GetInverseScaleFactors()

​

GetScaleSigmaSquares

std::vector<float> GetScaleSigmaSquares()

​

GetInverseScaleSigmaSquares

std::vector<float> GetInverseScaleSigmaSquares()

​

Public Members

​

mvImagePyramid

std::vector<cv::Mat> mvImagePyramid

​

Constants

​

Score Types

enum { HARRIS_SCORE=0, FAST_SCORE=1 }

• HARRIS_SCORE - Harris corner response
• FAST_SCORE - FAST corner response

​

Implementation Details

​

FAST Detection

1. Attempts detection with iniThFAST threshold
2. Falls back to minThFAST if insufficient features found
3. Ensures robust feature detection in various conditions

​

Octree Distribution

• Divides image into quadrants recursively
• Ensures spatial coverage across the image
• Prevents feature clustering in textured regions

​

Pyramid Levels

• Each level scaled by scaleFactor from previous level
• Features detected at all levels for scale invariance
• Typical configuration: 8 levels with scale factor 1.2

​

Usage Example

// Create extractor with 1000 features, 8 levels
ORBextractor extractor(1000, 1.2f, 8, 20, 7);

// Extract features from image
cv::Mat image = cv::imread("image.png", cv::IMREAD_GRAYSCALE);
std::vector<cv::KeyPoint> keypoints;
cv::Mat descriptors;
std::vector<int> lappingArea;

int nFeatures = extractor(image, cv::Mat(), keypoints, descriptors, lappingArea);

// Access pyramid information
int nLevels = extractor.GetLevels();
float scale = extractor.GetScaleFactor();
std::vector<cv::Mat> pyramid = extractor.mvImagePyramid;

​

See Also

• ORBmatcher - Feature matching using ORB descriptors
• Frame - Frame representation using ORB features
• KeyFrame - Keyframe with ORB features