Implementation of the fast plane detector as described in the paper "Fast Plane Detection and Polygonalization in noisy 3D Range Images", of Poppinga et al. (IROS 2008) https://ieeexplore.ieee.org/abstract/document/4650729
This module has the following dependencies:
- Eigen3
- Boost (array, thread)
- OpenCV
- ROS
This library can be used standalone without using ROS. In this way the module accepts a OpenCV depth image and outputs the detected plane. To use it in this mode you should use the following methods for include/plane_detector/plane_detector.hpp file:
-
PlaneDetector(double delta = 1.0, double epsilon = 1.0, double gamma = 10.0, int theta = 1000, double _std_dev = 0.03); The meaning of each parameter is available on the Popping paper. The std_dev is related to the accuracy of the sensor.
-
int detectPlanes(const cv::Mat &depth); (Input: depth image. Output: number of planes).
-
inline std::vector getPlanes() const; (gets the actual information of the detected planes)
To test the standalone version there is the test_plane_standalone.cpp example. It can read a depth image from the TUM dataset: https://vision.in.tum.de/data/datasets/rgbd-dataset and detects planes in that image.
Also, a testing ROS node is provided in src/test_plane_detector.cpp. In this case, the node subscribes to a depth stream from a RGB-D sensor (tested with OpenNI2 compatible devices) and generates markers and outputs the detected planes via stdout.
Future work should include the output of a specialized plane message for making it easier to use it in other modules. Any help is appreciated!!