yincanben / find-object Goto Github PK

Email from Florian Mertens:

"[...] The one thing I would find very useful, is if the device was a video 
file, that you could have a few buttons below the frame viewer for better 
browsing. Buttons such as: Play/Pause, next/previous frame, playing speed, a 
timeline scrollbar?"

Todo: 
1-Verify if it is possible to do that with actual OpenCV properties : 
http://docs.opencv.org/modules/highgui/doc/reading_and_writing_images_and_video.
html#videocapture-set

2- Modify Camera class, and add control Widget.

The problem is when a parameter is changed, this affects only the features' 
extracted in the scene, but not in added objects. If a different type of 
descriptors is used, the matching algorithm is disabled (a MessageBox warns the 
user). At this time (r73), the user has to manually delete/add the objects 
again. 

Todo (ideas):
- Add button "Update objects" under the Objects dock widget. The behavior is to 
remove features from the object's image and to re-extract them with the current 
parameters.
- This adds an issue, when adding object using the interface "Edit->Add 
Object...", on how the object region is defined. After stopping the camera, a 
new choice (using a combo box) could be added to select a rectangle (ROI) 
instead of selecting features. The features would be extracted from this region.
- When closing the application, notify the user if the objects were updated, to 
ignore changes or save objects again... 

Your utility was so useful for me getting started with feature detection / 
description. Any chance you will update it for the Agast detector and 
BRISK/FREAK descriptors? Even if not, thanks very much for sharing the utility.

Make it general to add new services in the future.

What steps will reproduce the problem?
1. Download Find-Object-0.4.6-Darwin.dmg
2. Move to /Applications, replacing existing Find-Object.app
3. Observe error on launch below due to libOpenNI.dylib not being found.

Find-Object 0.4.1 works fine.

Process:         Find-Object [7283]
Path:            /Applications/Find-Object.app/Contents/MacOS/Find-Object
Identifier:      Find-Object
Version:         ???
Code Type:       X86-64 (Native)
Parent Process:  launchd [181]
Responsible:     Find-Object [7283]
User ID:         501

Date/Time:       2014-06-30 21:43:42.573 +1000
OS Version:      Mac OS X 10.9.3 (13D65)
Report Version:  11
Anonymous UUID:  CB91B849-BD4F-C26C-CB2B-48D3904A50BD


Crashed Thread:  0

Exception Type:  EXC_BREAKPOINT (SIGTRAP)
Exception Codes: 0x0000000000000002, 0x0000000000000000

Application Specific Information:
dyld: launch, loading dependent libraries

Dyld Error Message:
  Library not loaded: libOpenNI.dylib
  Referenced from: /Applications/Find-Object.app/Contents/MacOS/libopencv_highgui.2.4.dylib
  Reason: image not found

Binary Images:
    0x7fff5fcdb000 -     0x7fff5fd0e817  dyld (239.4) <042C4CED-6FB2-3B1C-948B-CAF2EE3B9F7A> /usr/lib/dyld

The only way now to get detected objects and position outside Find-Object is 
with ROS. 

Since many users are on Windows/Mac OS X/ Ubuntu (without ROS), an interface to 
a Qt TCP socket could be used.

There are many parameters under the Feature2D panel... so only show parameters 
of the current selected detector/descriptor. 

On a mouse over on a parameter, show a tooltip to know what this parameter does.

What steps will reproduce the problem?
1../find_object
2../findobject-example

What is the expected output? What do you see instead?
Properly executed file

What version of the product are you using? On what operating system?
Newest Version - .0.4.1

Please provide any additional information below.
*** Error in `./find_object': realloc(): invalid old size: 0x0000000000697f60 
***
when ./find_object attempted
for example its the same thing but longer

Add a fixed vocabulary mode. Maybe by adding a parameter 
"General/FixedVocabularyPath" linking to a vocabulary file (like a session 
file).  

Refer to 
http://answers.opencv.org/question/547/using-flann-with-binary-descriptors-brief
orb/

Tried with the current OpenCV binaries on Ubuntu but the application crashes 
creating the index with matrix of type CV_8U. Retry when the next release of 
OpenCV comes out...

i have use your surf code to demonstrate the algorithm in image matching for my 
final year project.
However, when i try to extract the keypoints and descriptors in a image using 
surf 
it give me a error thats '"ERROR : kpt=%lu != 
descriptors=%d\n,keypoints.size(),descriptors.rows"
The details of code is show below:

                // EXTRACT KEYPOINTS
        cv::FeatureDetector * detector = Settings::createFeaturesDetector();
        std::vector<cv::KeyPoint> keypoints;
        detector->detect(img, keypoints);
        delete detector;
        ui_->label_timeDetection->setText(QString::number(time.restart()));

        // EXTRACT DESCRIPTORS
        cv::Mat descriptors;
        cv::DescriptorExtractor * extractor = Settings::createDescriptorsExtractor();
        extractor->compute(img, keypoints, descriptors);
        delete extractor;
        if((int)keypoints.size() != descriptors.rows)
        {
            printf("ERROR : kpt=%lu != descriptors=%d\n", keypoints.size(), descriptors.rows);
        }

do your know why this happen??

I also have a problem in converting the data in data change. 
When the code 'data.convertTo(dataTree_, CV_32F,1,0);' a error also come out.

OpenCV Error: One of arguments' values is out of range (The total matrix size 
does not fit to "size_t" type) 


would u please help me to guess what problem here??
Thanks

The console version of find-object:
$ find-object -console
uses QApplication. This requires X11 when launching the application on a 
server. It should use QCoreApplication.

Ideas:
- Make a FindObject object which doesn't depend on QMainWindow or QWidget. 
MainWindow will include this object to do the actual processing.
- The ObjWidget class should also be split for Gui stuff and non-Gui stuff.

I tried the implementation of main.cpp of find-object code example provided on 
the http://code.google.com/p/find-object/#Code_example. There is a bug when 
using the ORB with the LSH index. Would you like to tell me how to solve this 
bug. And I use the visual 2008, opencv2.4.2, and Window xp.

What steps will reproduce the problem?
1. Choose SIFT as detector
2. Choose ORB as descriptor 
3. Start camera

What is the expected output? 
Detecting...

What do you see instead?
Application closes, error:

OpenCV Error: Assertion failed (dsize.area() || (inv_scale_x > 0 && inv_scale_y 
> 0)) in resize, file 
/Users/mathieu/Documents/workspace/opencv-2.4.9/modules/imgproc/src/imgwarp.cpp,
 line 1835

Version 0.4.6 on Mac OS X...

What steps will reproduce the problem?
1.Descriptor brisk
2.Detector brisk
3.

What is the expected output? What do you see instead?
Application hangs

What version of the product are you using? On what operating system?
Latest 

Please provide any additional information below.

See title.

Is it possible to dump the visual dictionary with all the objects indexed, and 
load it through the config file ?

The default use of find-object is with a webcam but this could be useful if we 
could load objects from image files (assuming there is only the object in the 
image file) and a scene from an image file. Changing parameters would update 
the features of the objects and the scene. The same behavior can be also used 
when the camera is stopped.

Ideas:
- Add action to load a scene from an image file.
- Add action to load an object from an image file.
- When the camera is not running and a parameter is changed, update the 
features of the current scene (AND the objects ???) -> If the user has a large 
number of objects, this could be inefficient if after each small changes, all 
objects are refreshed. The default behavior for updating objects could be the 
use of the button "Update objects" introduced in issue 2. One could also add a 
boolean parameter to know if the objects refresh is automatic or must be done 
manually. In the last case, a message could be wrote in the status bar saying 
that objects' features must be updated is the parameters has changed.

What steps will reproduce the problem?
1.  Build the program from source on Windows
2.  Run from within IDE
3.  Hit space

What is the expected output? What do you see instead?
Expected output is the video stream, instead the app throws an error (see below)

What version of the product are you using? On what operating system?
0.4.0 and 0.4.1

Please provide any additional information below.

It looks like the config file is not being generated or not being completely 
generated.  The app crashes on line 125 of Camera.cpp, after tries to read the 
media path from the config file, which is a null string.  I can't figure out 
how that config file gets generated, or what the media path should be.  

When I run "Restore all default settings", the application crashes on line 1171 
of MainWindow.cpp,after a string error trying to read the kFeatures2D 
descriptors from Settings.  I think it's the same bug.

Thanks!

When resizing the objects, it they become too small, the colored features are 
not shown. So it is hard to know which homographs rectangle shown in the scene 
refers to which object. So find a way to show which color is the object... 
maybe like adding a colored rectangle around the object...

It would be great to change the resolution of the pictures received by the 
webcam as it would help to recognize low resolution pictures (eg.320x400).

Cheers,
Luc Michalski

ASIFT : http://demo.ipol.im/demo/my_affine_sift/

Python example from OpenCV:
https://github.com/Itseez/opencv/blob/master/samples/python2/asift.py

As for TCP, it would be really interesting to use the console version and 
getting the object info detected.

Cheers,
Luc

Is it possible to improve the project by multithreading the code? The process  
slow down tremendously when loading a large number of objects ! I think moving 
the Mainwindow:update function on more threads (i.e. one thread every 10 
objects) could  reduce the processing time! Any tips? Thanks!

When there are multiple occurrences of the same object, the current 
implementation will only detect one occurrence with the homography, considering 
the features of the other object all outliers.

Add an option to recompute the homography without inliers of the previous 
homography computed, to detect other similar objects, until there is no more 
accepted homography found.

For ROS, we should be able to send multiple homographies of the same object.

What steps will reproduce the problem?

1.compare the time of extraction of the BRISK descriptor with any other
2.
3.

What is the expected output? What do you see instead?
a time much lower, i see 400 ms

What version of the product are you using? On what operating system?
Find-Object-0.4.4 - Linux Ubuntu 12.04 LTS

Please provide any additional information below.

the execution time of extraction of the descriptor BRISK is inexplicably higher 
than the other binary description

Request Multiple Instance Recognition

By default, Find-Object (as well as the ROS node) loads parameters from the 
configuration file located in ~/.find_object/config.ini (Unix) or 
$USER_DIRECTORY$/Documents/FindObject/config.ini (Windows). So it uses always 
the last parameters set from the GUI.

TODO:
Add option in the Menu to save/load parameters from a "*.ini" file.

For the ros node:
Add parameter for find_object_2d node to specify a configuration file.

What steps will reproduce the problem?
1.select SURF as detector
2.select ORB as descriptor
3.

What is the expected output? What do you see instead?


What version of the product are you using? On what operating system?
Find-Object-0.4.4 - Linux Ubuntu 12.04 LTS

Please provide any additional information below.