steevo87/thermalvis

Cross-platform, OpenCV-based functionality for image processing and computer vision in thermal-infrared

thermalvis's Issues

Get it added to all recent ROS distros

May need to test it on several flavors e.g. Fuerte, Groovy, Hydro, Indigo?

Add a calibration example video to wiki

Add Windows XML ability to handle multiple nodes of the same type

Have an array of pointers for each node, and allow multiples of each type (up to a reasonable limit) to be parsed from the XML. Will need to enable user to link them together (perhaps using topic labels like within ROS?)

Optional launch of GUIs in non-ROS mode

A qt GUI should only be launched in non-ROS mode if a corresponding "reconfigure_gui" node (or whatever the modern ROS equivalent is) is specified in the launch file.

Prevent feature detectors from "panicking"

It seems that for the GFTT and HARRIS detectors, if NO features make the minimum sensitivity threshold, then they will effectively allow a whole heap of weaker features to make it through!

Extend input image processing modes

At the moment, it assumes 16-bit images have a conversion to temperature of T = G/100, which is fine, but you should add other default modes (e.g. for Miricle 307K), and also allow the user to configure a linear conversion between graylevels and temperature by providing the gradient and intercept.

Ensure Qt compiles properly in Linux

It should be ignored for ROS, because ROS has it's own interfaces defined.

Suppress CMake warning: Policy CMP0020 is not set

Occurs when the Generate button is pressed in CMake. Tried command <cmake_policy(SET CMP0020 NEW)> but it doesn't seem to do anything..

Remove qt-related Build warning

Occurs numerous times. Related to QT code:

C:\Qt\Qt5.3.1\5.3\msvc2012_opengl\include\QtCore/qchar.h(453): warning C4561: '__fastcall' incompatible with the '/clr' option: converting to '__stdcall'

Image processing defaults

Settle upon good default image-processing settings for both temperature-corrected (Optris PI450) and uncalibrated (Miricle 307K) video sequences that result in a similar level of contrast, and therefore enable the same default feature tracking settings to be used.

Option to output feature tracking data for analysis

Optionally output all feature locations, tracks and frame info to a file for MATLAB/other analysis.

Move toward QThread implementation

Cannot run Qt Widget GUI from non-main thread, so this sort of approach is necessary.

http://www.qtcentre.org/wiki/index.php?title=Updating_GUI_from_QThread

Add publications and citation information to wiki

Full output of per-frame feature detection information

Into TXT files, so that they can be analysed in MATLAB.

Implement optional GUI for changing flow parameters

Prevent auto-inclusion of Qt libraries in 64-bit mode

Currently cannot Generate properly, so best to prevent this behaviour by default.

Fix attemptMatching()

Returns many very poor matches in low SNR environments. Responsible for a lot of wonky tracks.

Add warning if in Qt mode it can’t find auto-generated header

Can then tell user to build the project using Qt Creator.

Ensure flow node fully working in ROS

Add ability to launch only streamer from monocularSLAM app

Improve the README formatting on main page

Maybe follow the wiki syntax

Add "fromFile" feature detection mode

Allow C++ feature detection to be bypassed in favour of using existing feature detection results stored in a directory. This is for the purpose of experimenting with external feature detection algorithms such as those implemented in MATLAB.

Validation of track recovery

Make sure track recovery works well on test sequences that include occlusion.

Correct initialization of real-time variables

Make sure initial state of qt variables is determined by provided XML values.

Perhaps should define the Qt form class to have some kind of constructor/initialization function which accepts a “streamerSharedData” datatype.

Should this be added to the Qt project? Or can you have a local wrapper function that simply passes all of the members of streamerSharedData to the Qt object, but doesn't require shared definitions or anything.

Get OpenCV debug libraries (PDBs) to load properly

They are being copied into the binary directory, but VS isn't loading them automatically.. maybe they need to go to another location?

Enable input parameters for feature tracking to be passed in via XML

Maintain identical process to ROS project. Include calibration data.

Validation of NUC/Interruption handling

Make sure NUC handling works when duplicate frames are detected.

Also, consider retroactively performing NUC handling if very few features are tracked in a new image (implying that there has been a significant transformation compared to the previous image, e.g. due to a looping sequence, missing frames, a NUC without duplicate frames stored etc).

Get system compiling and working again without Boost libraries

Obviously no directory reading or XML processing though.

set PCL to false if it thinks it has the wrong (32/64) library

develop a 64-bit solution for using Optris in real-time

At the moment the provided Optris SDK only seems to be able to compile in run in a 32-bit solution..

Integrate PCL with ROS implementation

If it can use an implementation already in ROS, that is preferred.

Implement XML input for streamer

Exception issue while running

Enable compilation and execution without boost libraries

At some stage this has been broken. Of course, features will be limited, but it should still compile and be able to run e.g. with default settings and the live device.

transform OptrisWinTest into OptrisWinSLAM

This Windows-only demo app should follow a similar design to the monocularSLAM app, but work with the live Optris, and perhaps utilize Windows forms instead of qtForms (eliminating the need for a Qt dependency).

Verify streamer performance in ROS

First it should just be able to read provided sample image data. Then verify it reading in and writing from a past rosbag. Finally ensure that it still works with the Optris ROS package (passing raw image data through to streamer to test).

Reduce the need to build each Qt project independently

Is there a way to combine them all into the one directory, so that the procedure only has to be executed once?

It shouldn't matter if the "main" function doesn't launch all the forms, or even any of them.

Ensure boost can be found in Linux/ROS

It's possible ROS will include this already (then great!)

Add prevention of Qt in 64-bit mode (incompatibility)

Hopefully later on you can get it working again and can re-enable this.

Improve adaptive-windows / velocity-prediction

Make sure you don't get any "gaps" in test sequences that cause failure of the tracking due to poor velocity prediction or adaptive windows.

Get everything working in Linux independent of ROS

Implement optional GUI for changing streamer parameters

Equivalent of ROS streamer GUI features. Want to avoid managed windows - perhaps use qt instead so that it can run on a non-ROS linux implementation as well?

Consider moving mesh reconstruction to an independent thread

Since it can be very computationally expensive, consider moving most of this (at least the slowest part) to its own thread. The growing point cloud can be rendered, but the mesh will only be superimposed periodically, when the algorithm finishes.

Verify installation and compilation on Linux with ROS

install openCV + ROS Hydro, get thermalvis compiling, and record experience in readme

Migrate and update demo launch files

Specifically those relating to streaming and optical flow. Should be both ROS and Windows compatible.

Add function to transfer Qt GUI variables to main program

Call it “updateRealtimeConfigurables()” or something. It should pass on changes from the real-time GUI to the parameters that are used by the actual nodes.

Put all thermalvis functionality in a namespace

May be able to use a declaration like this before you declare a class, which should put all of the class functionality in the namespace.

namespace thermalvis {
class some_class;
}

Enable output of debug-rendered tracking videos

So that you can review the detected and tracked features visually offline, moving through frame-by-frame.

Restore XML defaults, verify ROS-equivalent performance for feature tracking

Get Qt functionality working in 64-bit mode

Does it just have to do with getting the right installer? Or are there no appropriate installers, and instead you will have to compile from source to get this to work?