Features
This project contains 4 core ROS modules, and can be installed as a ROS package here.
- calibration
- data streaming
- sparse optical flow
- monocular SLAM (under restoration)
In addition, several stand-alone apps are available that can be compiled in either Windows or Linux, without needing ROS:
- dd-evaluator
- mm-calibrator
- monocularSLAM
Multi-modal and multi-device camera calibration system.
Built upon the OpenCV libraries, this camera calibration system is designed to be simpler and quicker to use than other alternatives - especially for the case of thermal-infrared cameras. It utilizes the latest code available in the OpenCV libraries for tasks such as pattern searching, distortion modelling and intrinsic and extrinsic calibration. It also includes an implementation of an optimal frame selection algorithm which eliminates the need for the user to manually select frames for calibration. Additionally, in order to facilitate simultaneous multi-modal calibration (such as of thermal-infrared and visible spectrum cameras), it includes an alternative pattern finding algorithm designed to accurately locate a calibration mask. The system can handle image sequences, or stored or streamed video as input.
The mm-calibrator system has also been integrated into ROS (Robotics Operating System).
On the left is a colorized thermal-infrared image of a reverse-cycle refrigeration external unit. On the right is the corrected image after the application of my calibration method. The lens distortion is almost completely removed, resulting in true straight lines appearing straight in the image.
The top row shows three images captured from a multi-device capture rig, comprising a thermal-infrared cameras surrounded by two conventional cameras. The bottom row shows the images after the application of my algorithms. Not only is lens distortion removed, but the views are all aligned vertically, allowing stereo-vision algorithms to be applied efficiently and accurately. The grid-like pattern held by the person (me) in view is part of the method I developed that has proven to be more effective than the conventional approach.
Multi-modal feature detector and descriptor evaluation system.
Built upon the OpenCV libraries, this evaluation system is designed to be more automated and comprehensive than conventional systems for evaluating local feature detectors and descriptors. It includes components such as detector sensitivity analyses, within-level transformation analyses and the ability to evaluate in and between different modalities (such as thermal-infrared). It also offers the ability to average detector and descriptor performance over a sequence of images, rather than just a single pair of images. The project is designed to be compiled and utilized easily provided that OpenCV is installed.
The two figures below show some of the typical output able to be produced by this system.
