Motion tracking algorithm
Locked
 
|
Hello
There is some kind of web cams that tracks the moving object and making it into the focus of its vision field. So it tracks the body . I really do not understand what is the algorithm of image processing in these types. I know that for motion detection, I should subtract a frame from the previous, so I get a new image that contains only the moving object, but what is the case if the cam itself moves? How can it know that the centroid of the moving object is into its focus ?? Actually I do want to make similar application using ordinary webcam based on 2 motors and external controller, but imagej will be the decider to say if the moving object is into focus or not. Can anyone help me regarding this? |
|
|
I have played with a number of motion detection programs, though I haven't tried tracking (PTZ minus Z) All the ones I have tested leave a lot to be desired. Part of the problem is CCDs will have some noise, so frame by frame comparison is not as easy as you think. Some average frames to reduce noise. They all have a slider to set the sensitivity.
My goal was not security but meteorite imaging. The camera would be in a fixed position and would save frames when motion was detected. Passing aircraft capture would be fine too. There is one rather expensive program for this task, but I figured I would test the free and low cost ones first. I think frame subtraction and some sort of "energy detect" on a grid is a start, but I don't think it will be satisfactory. Perhaps a frame to frame correlation analysis? Check out the VLC player, AKA videolan. They have a tracking plugin on that detects objects moving in the image. I think the plugins are open source. If you get serious about this, please start a sourceforge entry or some other place to store code. I'm certainly willing to beta test. Well assuming I can interface my frame grabber to imagej. ;-) ------Original Message------ From: mamech Sender: ImageJ Interest Group To: [hidden email] ReplyTo: ImageJ Interest Group Subject: Motion tracking algorithm Sent: Jun 10, 2011 5:36 PM Hello There is some kind of web cams that tracks the moving object and making it into the focus of its vision field. So it tracks the body . I really do not understand what is the algorithm of image processing in these types. I know that for motion detection, I should subtract a frame from the previous, so I get a new image that contains only the moving object, but what is the case if the cam itself moves? How can it know that the centroid of the moving object is into its focus ?? Actually I do want to make similar application using ordinary webcam based on 2 motors and external controller, but imagej will be the decider to say if the moving object is into focus or not. Can anyone help me regarding this? -- View this message in context: http://imagej.588099.n2.nabble.com/Motion-tracking-algorithm-tp6464200p6464200.html Sent from the ImageJ mailing list archive at Nabble.com. |
|
|
Sorry, but frame comparison is easy as I think : ) I have tested it using a program that I wrote it in java, and it worked. I have tested the plugin of the vlc player as you said and I have 2 problems with it: 1- I think it also uses frame comparison algorithm, so if both the cam and the target moves, then this algorithm will feel as if everything in the scene moves. Actually there are better programs for this than vlc I think. 2- This is from vlc forums: "You use it by adding "--video-filter motion" to your command line and it'll draw white rectangles around the moving areas in the video. The algorithm might be a bit buggy and i sure would appreciate if someone could test it some more and provide feedback." I am quiet sure that there is smoe algorithm used for motion detection of certain object when both target and camera are moving |
|
|
Frame comparison works for very large moving objects, but can't track
small moving objects. Once you crank up the sensitivity, noise is the culprit. It would probably pay to do a good search to see if this code exists since there are at least 10 programs I've come across that do simple frame comparison. None met my needs, which may be more difficult than what you are trying to accomplish. I uninstalled most of them, but kept http://www.yawcam.com/ http://www.trackercam.com/TCamWeb/index.htm Handiavi is supposed to handle meteor detection, but I didn't find it much better than the other programs. > http://www.azcendant.com/Meteor%20Capture.htm I had no luck getting Zoneminder to install, but that is a well known motion detection program. Of the papers published, most use some variant of the Hough transform. On 6/10/2011 7:24 PM, mamech wrote: > Gary Sellani wrote: >> >> I have played with a number of motion detection programs, though I haven't >> tried tracking (PTZ minus Z) All the ones I have tested leave a lot to be >> desired. Part of the problem is CCDs will have some noise, so frame by >> frame comparison is not as easy as you think. Some average frames to >> reduce noise. They all have a slider to set the sensitivity. >> > > Sorry, but frame comparison is easy as I think : ) I have tested it using a > program that I wrote it in java, and it worked. > > > > Gary Sellani wrote: >> >> >> Check out the VLC player, AKA videolan. They have a tracking plugin on >> that detects objects moving in the image. I think the plugins are open >> source. >> >> If you get serious about this, please start a sourceforge entry or some >> other place to store code. I'm certainly willing to beta test. Well >> assuming I can interface my frame grabber to imagej. ;-) >> >> > > I have tested the plugin of the vlc player as you said and I have 2 problems > with it: > > 1- I think it also uses frame comparison algorithm, so if both the cam and > the target moves, then this algorithm will feel as if everything in the > scene moves. Actually there are better programs for this than vlc I think. > > 2- This is from vlc forums: > "You use it by adding "--video-filter motion" to your command line and it'll > draw white rectangles around the moving areas in the video. The algorithm > might be a bit buggy and i sure would appreciate if someone could test it > some more and provide feedback." > > > I am quiet sure that there is smoe algorithm used for motion detection of > certain object when both target > and camera are moving > > -- > View this message in context: http://imagej.588099.n2.nabble.com/Motion-tracking-algorithm-tp6464200p6464373.html > Sent from the ImageJ mailing list archive at Nabble.com. > |
|
|
In reply to this post by mamech
does mtrack2 do what you want?
http://valelab.ucsf.edu/%7Enico/IJplugins/MTrack2.html chris On 11 Jun 2011, at 01:36, mamech wrote: > Hello > > There is some kind of web cams that tracks the moving object and making it > into the focus of its vision field. So it tracks the body . > > I really do not understand what is the algorithm of image processing in > these types. I know that for motion detection, I should subtract a frame > from the previous, so I get a new image that contains only the moving > object, but what is the case if the cam itself moves? How can it know that > the centroid of the moving object is into its focus ?? > > Actually I do want to make similar application using ordinary webcam based > on 2 motors and external controller, but imagej will be the decider to say > if the moving object is into focus or not. > > > Can anyone help me regarding this? > > -- > View this message in context: http://imagej.588099.n2.nabble.com/Motion-tracking-algorithm-tp6464200p6464200.html > Sent from the ImageJ mailing list archive at Nabble.com. |
Re: Motion tracking algorithm
|
|
In reply to this post by mamech
Hi,
did you check out the (Extended) Kalman Filter? - Stephan Am 11.06.2011 um 02:36 schrieb mamech <[hidden email]>: > Hello > > There is some kind of web cams that tracks the moving object and making it > into the focus of its vision field. So it tracks the body . > > I really do not understand what is the algorithm of image processing in > these types. I know that for motion detection, I should subtract a frame > from the previous, so I get a new image that contains only the moving > object, but what is the case if the cam itself moves? How can it know that > the centroid of the moving object is into its focus ?? > > Actually I do want to make similar application using ordinary webcam based > on 2 motors and external controller, but imagej will be the decider to say > if the moving object is into focus or not. > > > Can anyone help me regarding this? > > -- > View this message in context: http://imagej.588099.n2.nabble.com/Motion-tracking-algorithm-tp6464200p6464200.html > Sent from the ImageJ mailing list archive at Nabble.com. |
|
|
In reply to this post by chris elliott
Sorry, but after I tried it I could not understand its main function. Are not there any tutorial for it ? |
|
|
In reply to this post by Stephan Preibisch
Forgive me because my knowledge in image processing field is limited, but what is the relation between kalman filter and object detection ? my main problem is not the noise. |
|
|
In reply to this post by mamech
Noise is always an issue with video. Even if the CCD was perfect, there are issues due to thermals, tree branches move, shadows from items out of the view of the camera, etc.
Like I said, these simple frame comparison trackers are quite poor when operated in the real world unless the sensitivity is very coarse. ------Original Message------ From: mamech Sender: ImageJ Interest Group To: [hidden email] ReplyTo: ImageJ Interest Group Subject: Re: Motion tracking algorithm Sent: Jun 11, 2011 10:06 AM Stephan Preibisch wrote: > > Hi, > > did you check out the (Extended) Kalman Filter? > > - Stephan > > > Forgive me because my knowledge in image processing field is limited, but what is the relation between kalman filter and object detection ? my main problem is not the noise. -- View this message in context: http://imagej.588099.n2.nabble.com/Motion-tracking-algorithm-tp6464200p6465652.html Sent from the ImageJ mailing list archive at Nabble.com. |
|
|
Noise will be an issue after I know how the backbone of the detection algorithm should be in the mentioned case Anyway, this depend on the application itself. For me it satisfactory. |
|
|
mamech,
OpenCV software may be worth looking at see http://www.youtube.com/watch?v=bWl33urh2w8 and suggestions On Sat, Jun 11, 2011 at 6:46 PM, mamech <[hidden email]> wrote: > Gary Sellani wrote: >> >> Noise is always an issue with video. Even if the CCD was perfect, there >> are issues due to thermals, tree branches move, shadows from items out of >> the view of the camera, etc. >> > > Noise will be an issue after I know how the backbone of the detection > algorithm should be in the mentioned case > > > > Gary Sellani wrote: >> >> >> Like I said, these simple frame comparison trackers are quite poor when >> operated in the real world unless the sensitivity is very coarse. >> > > Anyway, this depend on the application itself. For me it satisfactory. > > > -- > View this message in context: http://imagej.588099.n2.nabble.com/Motion-tracking-algorithm-tp6464200p6466141.html > Sent from the ImageJ mailing list archive at Nabble.com. > |
|
|
Thank you very much
But It uses the color of the object for detection, also his camera do not move to make the object in its focus. I began to understand where the problem is, but there is something that I want to ask about: If I have binary image showing 2 different areas (each area represents an object in real life), and I want to make a java code (based on imagej) to get the area and the centroid of each one. I tried to make this (with recording macros) by using create selection, then set measurement, but it seems that imagej deals with separated objects as one object. Any ideas about how to do so ? Thanks |
|
|
If you were doing this using the menus, then I think Analyze/Analyze
Particles would give you a Results Table with this info in it. Note, you must use Analyze/Set Measurements to define what blob descriptors you actual want (e.g. Centroid). You can get this as a Java plugin by using the macro recorder to record the Analyze/Analyze Particles step and then use the record and create buttons to convert that to a plugin. Hope this helps David Webster On Mon, Jun 13, 2011 at 7:45 PM, mamech <[hidden email]> wrote: > Thank you very much > > But It uses the color of the object for detection, also his camera do not > move to make the object in its focus. > > I began to understand where the problem is, but there is something that I > want to ask about: > > If I have binary image showing 2 different areas (each area represents an > object in real life), and I want to make a java code (based on imagej) to > get the area and the centroid of each one. I tried to make this (with > recording macros) by using create selection, then set measurement, but it > seems that imagej deals with separated objects as one object. > > Any ideas about how to do so ? > > Thanks > > -- > View this message in context: > http://imagej.588099.n2.nabble.com/Motion-tracking-algorithm-tp6464200p6472568.html > Sent from the ImageJ mailing list archive at Nabble.com. > |
«
Return to ImageJ
|
1 view|%1 views
| Free forum by Nabble | Edit this page |