Prasanna's profile - activity

Oh. I didn't see the date :D Maybe I will try mashing it up and if something comes up I will let you know.

Maybe Rect res is Rect(0,0,0,0) as _last_guess might have converged to 0,0,0,0. You should probably check that to confirm.

I still don't get why you need to add a specific rule (swarm / hog) in the implementation of particle filter. Won't the distribution by itself take care of that? The particles might be unstable but the mean estimate of the state need not be that unstable. This blog seems to have a strategy for resampling and regularizing.

Are you still maintaining this code elsewhere (github, bitbucket, etc.)? If so, can you post a link here? Has it been updated after this?

Well that was an amazing video indeed. Thanks. I came across this kind of sensors under the name of Neuromorphic Cameras and Optical Flow sensors. Pretty amazing things they are.

Regarding S3E, The technique they used is based on a concensus transform or something - Never heard of it before. Will share any details if I come across anything interesting.

Well, Maybe you should wait for Kinect 2.0 with which you can experiment with Time of Flight for depth. There are these techniques which I came across - Depth from blur, Depth from defocus, Depth from Photometry - Don't know anything about their performance, just heard about them. Maybe you wanna look into them.

What I actually meant by curvature was - When you set your stereo rig in front of a curved surface like a cylinder / sphere, Can you get a depth map such that the depth will vary as you move along the surface or will it just be like a depth map you get from repeating the experiment with a cube?

In short - Will you be able to differentiate a cube from a cylinder of same dimensions - side of cube = height of cylinder = dia. of cylinder?

Well, I haven't come across any real time stereo algorithm working without any additional hardware like GPGPUs etc. The video link you posted was indeed a surprise to me - How could they possibly do it? I will dig a bit more. Thanks for sharing! I know you would have come across devices like kinect (Infrared ) and Time of Flight Cameras. Why don't you give them a try too? They seem to work real time. And regarding curvature, I didn't get your point on how HoG helps. Can you please elaborate? I will be happy even if someone can point out if its impossible to get the curved surface of a cylinder / sphere with a stereo rig.

Check this out. https://github.com/MasteringOpenCV/code. Check chapter 4.

UPDATE

I think I got the problem solved. It was due to a destructor of a Global Mat being called after the program ends. Its funny that this should happen. I donno whether to call this a bug or not. I released it just before the return statement of main function and things are fine.

PS: If you want to replicate error, try using a Global Mat with just a simple "imread - imshow - waitKey " program.

.

Actual Question

I was running some OpenCV (3.0.0) codes today. I stumbled upon the following runtime error.

pure virtual method called

terminate called without an active exception

Aborted (core dumped)

This is my gdb backtrace.

(gdb) bt
0  0x00007ffff615bf77 in __GI_raise (sig=sig@entry=6) at ../nptl/sysdeps/unix/sysv/linux/raise.c:56

1  0x00007ffff615f5e8 in __GI_abort () at abort.c:90

2  0x00007ffff67636e5 in __gnu_cxx::__verbose_terminate_handler() () from /usr/lib/x86_64-linux-
gnu/libstdc++.so.6
3  0x00007ffff6761856 in ?? () from /usr/lib/x86_64-linux-gnu/libstdc++.so.6

4  0x00007ffff6761883 in std::terminate() () from /usr/lib/x86_64-linux-gnu/libstdc++.so.6

5  0x00007ffff676235f in __cxa_pure_virtual () from /usr/lib/x86_64-linux-gnu/libstdc++.so.6

6  0x00007ffff74d5d8c in cv::Mat::deallocate() () from /usr/local/lib/libopencv_core.so.3.0

7  0x0000000000405fdb in cv::Mat::release() ()

8  0x0000000000405d40 in cv::Mat::~Mat() ()

9  0x00007ffff6161071 in __run_exit_handlers (status=1, listp=0x7ffff64e66a8 <__exit_funcs>,   run_list_atexit=run_list_atexit@entry=true) at exit.c:77

10 0x00007ffff61610f5 in __GI_exit (status=<optimized out>) at exit.c:99

11 0x00007ffff6146dec in __libc_start_main (main=0x405451 < main >, argc=2, ubp_av=0x7fffffffe048, 
init= < optimized out > , fini= < optimized out >, rtld_fini = < optimized out >, stack_end=0x7fffffffe038) at libc-start.c:294

12 0x0000000000402f49 in _start ()

When I did some searching around, people pointed out some deallocation / access after lifetime and some were about threads. But I didnt get head or tail of it.

It would be nice if someone could give an explanation.

I think there is one more thing you can do about this. You can use the 'appsrc' element in GStreamer to do this. I haven't tried it out myself. But I just know its possible. Check it out if you are still exploring. :)

There is something called hierarchy in contours. Look it up. Maybe it can help you. A Naive way is area constraint. Just remove the 'contour' having highest area, has a 'hole' inside or something like that. (If you dont know what is the difference in 'hole' and 'contour', check the learning opencv oreilly book).

Hey Steve, Look what happened to your asterisks. Edit it before you confuse others. :D

What you can also try is to use std::vector of points. This will eliminate the malloc (did you forget to allocate memory in any case? ) step and problems associated with pointers. For example, std::vector<Point> points; and use push_back to add points to the vector, like this, points.push_back( Point(x,y) ); and you can access by points[i] where i is index.

Try to use vectors whenever possible (but sadly there are drawbacks too. :( But on the bright side, its way better and easy to handle than pointers)

Theres a waitKey(0) between showing samples from a vec file. Hit the enter key to go ahead. Make sure you merge the vecs created for each image.

You can open your "negative.dat" file in a text editor. As for verifying a vec file, Use the opencv_createsamples.exe by passing vec file, width and height as arguments and -show option. (For example, opencv_createsamples.exe -vec imgs/vector.vec -w 640 -h 480 -show). But why are you using such high resolution like 640x480? It will take ages to train.

Try following this for a start.

Wait a sec. I didn't point that out as a solution to the actual problem. My guess is to try passing paths with a " ./ " preceding it. So it should look like ./negative/UMD_001.jpg . Tell me if it works.

Parameter which you passed is "preCalcValBufSize" while it should have been precalcValBufSize. Similarly in the case of "precalcIdxBufSize". Its a sad thing that the parameters are not checked by the program. Double check your bg file too.

If you need more help, contact me at [email protected]. I can share my experience with cascade classification with you

My guess is the data chosen for training is not correct. The training is suffering too much from the variance in the data set and trying to add more features to discriminate between them and ending up being a BIG FAT cascade. Please Note that Its a guess.

And I really want to appreciate your idea. Nice work.

And i would suggest that instead of making a very robust detector, make one for the situation at hand. Take natural positives and negatives from the scene where it is used. You can make a cascade with few images, but you wont be able to capture all possible 'test' cases with a small dataset.

One more thing. You can also try other feature evaluators other than LBP like HOG,HAAR etc. You can also try SIFT, SURF, etc. too.

I am really sorry, there are back quote marks (below tilde symbol) marks before and after pkg-config. So the command is g++ -o facedetect facedetect.cpp(back quote)pkg-config --cflags --libs opencv(back quote)` My bad.

And why do u need relative path for PKG_CONFIG_PATH? Use absolute path of opencv.pc in export command.

I assume that the quote mark ( ' ) at the end of your previous comment (cv::CascadeClassifier::CascadeClassifier()') was a mistake. All I wanted to know was whether your opencv has been built correctly or not as these errors can come because of that too. Try compiling and executing some other program and test if opencv works.

pkg-config takes care of all the linking provided its used correctly. Verify whether the path is set by using : echo $PKG_CONFIG_PATH command in terminal.

For compiling, g++ -o facedetect facedetect.cpp pkg-config --cflags --libs opencv

One more thing that you can do is edit the CMakeCache.txt generated after running cmake command without any parameters. Otherwise, You can also use the CMake GUI and select the advanced check box to see all the optional Parameters. Hover your mouse over the options to get a small description about the parameters.

I dont think I understood your requirements completely. But, As far as I understood, you wont touch CUDA code as long as your are using OpenCV for your application as it is all wrapped up into the gpu module. (its kinda sad as the opportunity to learn CUDA goes down :( ).

If you want to use CUDA code, you need to get to the base of the streamer (for grabbing video frames) and convert it into the format which your code can accept. Looks tough (I havent tried it though). Otherwise, OpenCV has all the major things you will need. (Edge detection -> Canny, Template matching -> gpu::matchTemplate)

You can process at a greater FPS on videos, but not on cam as you need to upload and download data everytime (to the GPU) which results in lower fps.

In the folder C:\cygwin\tmp\opencv-2.4.5\samples\c\, there will be a file facedetect.cpp. Try compiling it. ( If you have specified the 'build all examples' flag when you built OpenCV, you will find a facedetect.exe (or) a c-example-facedetect.exe in the bin or somewhere. Search for it.) Try running it. It is just to make sure your OpenCV and its utilities are built properly.

Hey there. Can you please let us know if you are able to run other samples like facedetect, etc.?

Not sure if the error has been rectified or you got it all wrong :| .

Dilate -> Local Maximum. Erode -> Local Minimum.

It depends on what you passed as the command line argument. If it is the relative path, the picture must be in the folder having the executable generated by building the project. If it is the absolute path, well, that's a weird error.

Its really awesome that you made your own tool. Let me clear the air about opencv_createsamples. See if it helps.

You need to use the opencv_createsamples utility on each of the positive images (you have to pass image and not info.dat) and create 5 new images with different (random) distortions (perspective and intensity). You can skip bg file (A black background will be used if you do). This will create a vec file for each of the image with the distortions. Then you should use the mergevec.cpp to merge these vecs into one. Kindly look at the tutorial for more info about this merging of vec files. (You will have 109 vec files and 109*5 positive samples if you use -num 5)

Note : These images created are not a perfect substitute for the lack of original images. They seem to work but they cant capture all of the "variance" in the positive class. It will work the best when you have 600 natural positive images.

So,

• Use createsamples utility to generate the vec file for each image with random distortions
• Merge these vec files using mergevec.cpp provided here along with instructions.
• Put the path of negative images (relative to bg.txt) in bg.txt
• Pass the final (merged) vec file, bg.txt, and the rest of the parameters (like feature type, stages, falseAlarm Rate, HitRate, etc.) to traincascade to train.

But once again, It was really nice that you put time into it and solved it in a different way.

If you need any more clarifications, comment below. Glad to help.

Regards,

Prasanna S

PS : We need that guide soon, @steve. People are losing hope on opencv_createsamples utility :( . Hope the developers' feelings aren't hurt.

Actually, It will be very helpful to have a single guide because all the guides presently available online are sort of incomplete and the tutorial i mentioned is the best one i came across so far. I don't think you guys need help, but If you do, consider me in. I am not a pro but would love to help. Cheers!

Maybe you should look into ITK and its examples for a solution. (Donno if you will find something suitable. Just suggesting) It has OpenCV IO Bridge as well.