gnebehay / cmt Goto Github PK

License: Other

Python 100.00%

cmt's Introduction

Introduction

CMT (Consensus-based Matching and Tracking of Keypoints for Object Tracking) is a novel keypoint-based method for long-term model-free object tracking in a combined matching-and-tracking framework. Details can be found on the project page and in our publication. The Python implementation in this repository is platform-independent and runs on Linux, Windows and OS X.

License

CMT is freely available under the 3-clause BSD license, meaning that you can basically do with the code whatever you want. If you use our algorithm in scientific work, please cite our publication

@inproceedings{Nebehay2015CVPR,
    author = {Nebehay, Georg and Pflugfelder, Roman},
    booktitle = {Computer Vision and Pattern Recognition},
    month = jun,
    publisher = {IEEE},
    title = {Clustering of {Static-Adaptive} Correspondences for Deformable Object Tracking},
    year = {2015}
}

Dependencies

Python
OpenCV-Python (>= 2.4, < 3)
NumPy
SciPy
optional: ipdb (for debugging the code)

Note for Windows users: if you are unable to read video files, please follow this suggestion: http://stackoverflow.com/questions/11699298/opencv-2-4-videocapture-not-working-on-windows

Usage

usage: run.py [-h] [--challenge] [--preview] [--no-preview] [--no-scale]
               [--no-rotation] [--bbox BBOX] [--pause] [--output-dir OUTPUT]
               [--quiet]
               [inputpath]

Optional arguments

inputpath The input path.
-h, --help show help message and exit
--challenge Enter challenge mode.
--preview Force preview
--no-preview Disable preview
--no-scale Disable scale estimation
--with-rotation Enable rotation estimation
--bbox BBOX Specify initial bounding box. Format: x,y,w,h
--pause Pause after each frame
--skip N Skips N frames of the video input
--output-dir OUTPUT Specify a directory for output data.
--quiet Do not show graphical output (Useful in combination with --output-dir).

Object Selection

Press any key to stop the preview stream. Left click to select the top left bounding box corner and left click again to select the bottom right corner.

Examples

When using a webcam, no arguments are necessary:

python run.py

When using a video, the path to the file has to be given as an input parameter:

python run.py /home/cmt/test.avi

It is also possible to specify the initial bounding box on the command line.

python run.py --bbox=123,85,60,140 /home/cmt/test.avi

Use a sequence of numbered image files as an input:

python run.py sequence_dir/{:08d}.jpg

Here, {:08d} is a python format string that is expanded to 00000001.jpg, 00000002.jpg, etc.

cmt's People

Contributors

Stargazers

Watchers

Forkers

uping jayrambhia mapiarz splade colincsl amiltonwong kevleyski nagyist arthurbenemann nmathew87 lentin aahumada-git pirobot simedcn sendel dtmoodie colingogo galsont the13thson bubulemag bibongo tianwalker2012 se7oluti0n abramhindle jieguo noodlebreak chagge xiaozhuka azuredsky yuanboshe mayank-technical pianoza deraltefritz leizi007 praveenv4k physicsg anathan eljeilany samannp svmtracking maxint markstrefford rhayman ikrishneel ggaoming saberguo tuanthng cvml tanger830 toinsson rickesh wucanyi miaomiao1989 xianyuxiaobai spaceman007 longchuan1985 caomw golden1232004 jizhihang martiinsiimon laplizard bbetter173 dyzhou2015 aptrg haoyangwang ground-gcs olivierhub zhangzhensong mobcdi tachim heavenyc airyym gezhiwei aerobota sunzhusz lukacu xchenjh walkoncross bryant1410 sunkaianna yueyangming kuonanhong akshay-sn jorisgu vbillys essort yongjun823 gn02596703 shuchitagupta duyongkim sabarishkumaramaravadi jacoblee121 kulikovv jiapei100 hankerbit moshanatucsd pandinosaurus astoncpou shockzort alexadlu

cmt's Issues

Segmentation fault (core dumped)

I'm getting a core dumped fault on ubuntu 14.04 when I run the command
CMT$ python run.py challenge.mp4

challenge.mp4 is located in the CMT directory.

Any help with that is appreciated

Any plans to support OpenCV > 3

As opencv is now available at 3.1, are there any plans to upgrade CMT to work with 3.1
If not is there a particular sweet spot of OpenCV 3 that you found works ?

Undefined module/variable

When trying to execute lines the following lines ( by pressing 'd') the program has an exception.
https://github.com/gnebehay/CMT/blob/master/run.py#L219-L220

Here is the console output:

Traceback (most recent call last):
  File "run.py", line 215, in <module>
    ipdb.set_trace()
NameError: name 'ipdb' is not defined

Need help with an error.

I am constantly getting the error below. Can someone help? I have modified the code to run for multiple objects by initializing it with a new region to start with. Please help. It stops throwing this error at times.
Thanks.

File "run2.py", line 136, in
CMT.process_frame(im_gray)
File "/home/shuchita/Documents/Internship/CMT-master/CMT.py", line 245, in process_frame
selected_matches_all = self.matcher.knnMatch(features, self.selected_features, len(self.selected_features))
TypeError: object of type 'NoneType' has no len()

CMT not reading files

So the original issue was:
"I am trying to run CMT on some of the provided data sets and I am encountering the OpenCV error: Assertion failed (scn == 3 || scn == 4). Any Idea what could be causing this? seems to be called in run.py at line 104."
This seems to be alleviated by adding:
if len(im0.shape) == 3:
im_gray0 = cv2.cvtColor(im0,cv2.COLOR_BGR2GRAY)
else:
im_gray0 = im0
this in turn revealed what I think was the underlying issue. I keep getting that im0 is equal to None or None type meaning that either the program is not finding the specified files of the first frame is blank or "dead". I have looked around and have found some suggestions to try to fix this problem. Most add some code like :
" frame=cvQueryFrame(capture);
if (frame..empty()) break;"
(code from: http://stackoverflow.com/questions/21871540/opencv-error-assertion-failed-scn-3-scn-4?lq=1)

or:
if defects is None: else:

I am trying to fix it but being relatively new to Python I cannot get the program to rid the given video files, both color or gray scale.

cap = util.FileVideoCapture(args.inputpath) AttributeError: 'module' object has no attribute 'FileVideoCapture'

CMT for multiple objects

Can CMT be used for multiple objects in a frame? If yes, please suggest the best way that could be possible.

Question on Key points

When using sample files there seem to be some differences in the outputs shown than to my result. Number of key points is less in my outputs then in shown results. Key points are also tightly cluster in center of box, all in center are coloured blue. Any points not in exact center of Box are white. None are green like samples, why did this change, is this for newer versions? Also in results like cup on table, the cup is not tracked and is lost very easily by program. I can post output if needed. Sorry for bad english.

Python version not as robust as CPP version

I just ran some of my own datasets on the CPP version. It worked great, but it was very noisy (keypoints detected all over) in the python version. Why might that be?

Wrong DESC_LENGTH?

Hi Georg

The DESC_LENGTH in your Code is hardcoded 512.
But the dtype of the BRIEF (or ORB) descriptor is uint8 and the length is 32.
I think it should be 8 times 32 = 256 as written in BRIEF: Binary Robust Independent
Elementary Features in section 3.3 Distance Distributions

BR Clemens