View Full Version : Counting and tracking people from video
July 24, 2004, 10:07:00
i want to show a working concet for counting people using a video camera which view the entrance of a building in an angle from front and not from top.
i am planing to do so by background elimination. blob tracking and counting.
three main problems are
1. as i am shoting from front and need to know when did the person or people crossed specific distance which from my point of view equal an enterance.
2. how do i lock on more then one object
3. i would like to search for elipses as representing the human head, how to do it?....for ex. how can i teach montivision blob finder to search for an eliptical shape.
appreciate your reply
July 25, 2004, 20:43:43
I will start with point 3. We saved some predefined profiles which can be accessed thru the Load/Save property page/interface. You can train every kind of shape by feeding the blob finder with shapes and look at the resulting hu invariants. You should add some tolerances to your finally chosen hu ranges.
Point 2: We plan to change the blob finder to detect more than one blob, it will probably be a new modul like MV Multi Blob Finder.
Point 1: Not really a good idea how to solve that, maybe someone else can help ...
July 25, 2004, 22:27:38
1. Thanks for the moral support :)
2. At this stage i am only interesting in locking one person only.
3. How to i train the blob finder anyway through the load function? with an elipse shape for example?
4. For some reson, every time i do a "mistake" trying to connect one mv filter to another which can not be connected, the software breaks the layout and spread all icons seperatly....this is very annoying.
5. it seems to me through using your software i might be able to present a basic working model very quuickly, is there any way i can get your full working software under full students rights.....i am a student at TA university in israel
6. How can i take to bmp files, compare them, get an output which is a bmp picture showing only the difference betweem the two. Also i would like to know if i have for ex. the background picture as a reference how can i "delete" it in other pictures where is the same backgournd only with additional person in it....this way i can get only the object and let the system "blob" it much easier.
7. my final aim is to show a working concept of an avi movie in which a person walks in a coridor in front of a camera, the system must detect a momvment, lock the person head/body and starts measure it, once it croess an imgenary line which represents lets say a two meter distance, it means the object size got bigger up to the stage that i can assume it crosses that bar meaning i can count this event as an enterence.....any thoughts how to implement it
July 28, 2004, 16:55:55
3. Yes, for example. But add tolerance to the Hu Invariants in order to be more tolerant :)
4. Yes, sorry, we know the issue but havn't really found a good solution yet.
5. Would a time limited trial license do for the demo? Contact me at email@example.com to get a trial license. If you decide to purchase the software, we offer an educational version for use in a university department at half the price of a one developer version. This license would only be valid for educational purposes.
6. Connect the outputs of two MV Image File Source Filters to the MV Absolute Difference filter, you can capture the result using the MV Image Grabber. You can also combine one image and a live video source.
7. Yes, the size change is a possibility, but relative to which start size? Thats the question ...
July 28, 2004, 23:04:39
1. Thanks for your real reply...
2. about your offer of a limited trial license and that This license would only be valid for educational purposes. I fully agree, this is only for learning purposes, and in any case i have to finish my thesis in two month time so that limitd trial works great for me.....of course i will pass your development software capabilities to my supervisor for him to decide weather or not he would like to add this great tool of yours into his pool of lab tools.
3. About your explanation of working with the two MV file source filters and the MV absolue filter, i would like to use first image as background reference and the second mv as the video source, hence being able to substrack the bacground from the video and haveing only a b/w imagery of the person with black background, and from there to pass it to the blob finder....Can this work, please explain a bit more who to do it, as when i try it, i get funny results and unvalid connection in some cases like from the absolute filer to BW converter.
4. About your remark that the size change is a possibility, but relative to which start size? Thats the question ... well i thought of statistics measurments, meaning just for the idea itself that if a person head is X round with X pixels from a Fixed distance, then after two meters further to the camera, his potential size will grow around X precent, meaning an entarance.....It's not a scientific model, but a fast one that might work in most cases (i hope ).....love to know your thoughts.
Thanks for your time
July 29, 2004, 08:27:25
3. Did you make sure that the video and the reference image do have teh same dimension?
Video Capture Filter -->
MV Abs Diff -> Renderer
Image File Source -->
This (a bit abstract) configuration should do fine.
4. Sounds to be worth a try ...
August 19, 2004, 15:51:21
1. Congratulations on 2.2 release.
2. Following our corespondance and now that you know what i would like to show do you think ver 2.2 has most if not all the tools i need.
3. I do need your geedback about it and help as much as possible.
August 19, 2004, 22:27:59
Please write down which points you solved and on which you are still working in order to give me a base knowledge to start with.
August 20, 2004, 21:28:47
1. I have managed to use all filters and set up to get resonable single blob count and lock all over the video footage of a person walking in a corridor
What i do not know how to do and need you help is as follows:
1. How do use elipsses mask so the blob will try and lock only on the head shape. I saw in the new tools there is an option of loading an elpises but i have no idea what it actualy does.
2. How do i save counter information externaly and convert number to text in order to overlay it on top of the video.
3. how to estimate the size of the blob in every frame and add a condition that when it reaches a certin size to raise a counter by one, or reduce it for that matter.
I have attached for you the basic project file but i can not send you the 3 sample BMP files that i used as not enough space allowed.
Looking forward for your experties and assistance.
August 23, 2004, 10:28:02
1. It loads a set of Hu variants that were obtained from an ellipse. As i said before, you can load an ellipse image into the blob finder and display its Hu Invariants. Those values (plus some tolerance) can be used as contour filter.
2. You obtain the information thrue the filters COM interface (IMVBlobCounter (2)) and you can create a text overlay using the MV Text Overlay filter.
3. The new IMVBlobCounter2 interface delivers information about the detected blobs, e.g. position, area, ...
August 23, 2004, 14:42:39
Sorry to nag you BUT still i hope you understand my stress :(
1. About the Elipsses....this is all fine but my questions is for example, if i tracked a single blob containging human figure and only inside this cotour i would like to search for an elipse counter hence the person head, how do i do it?.....and another thing regarding this as the person is walking towards the camera how can i tell the software to adpot the change in figure size.
2. From your reply i uderstand that i will have to use a higher level software support such as VB or VC......Is this true or you the "IF....THEN....ELSE" will be part of the MV Tool in ther future than can hanlde integer for that matter.....could be a great simple idea to add in btw.
3. I hope you managed to look at my filter with the samples i send you....i would like to know your feedbacks and remarks how to preform it better.
Powered by vBulletin® Version 4.2.2 Copyright © 2014 vBulletin Solutions, Inc. All rights reserved.