Intelligent Traffic Surveillance

This video shows some results from our work on intelligent video surveillance. Multiple targets are tracked across the scenes in the busy streets of Kanpur. The tracking algorithm does'nt assume any prior object model but learns and use the same to track. The algorithm works in real time - both background subtraction and tracking are accomplished at 10 frames per second on 320 x 240 images on an Intel 3.0 GHz single processor PC. If you are interested in our research work, please contact me at prithwijit.guha@gmail.com

Category:

Entertainment

Tags:

• smart
• video
• surveillance
• multiple
• target
• tracking
• object
• avtivity
• recognition

[nuclearthreat545]

How many objects can it track at once?

[pguha77]

@nuclearthreat545

No Upper Limit like that. Higher the number of objects, slower the code is.

[ferchuhill]

First, thanks for answering. Second, I would like to know where can I get information about First Mean-Shift and then KL, some paper in particular?

[pguha77]

For mean-shift tracking...Refer to the paper in CVPR2000 by Comaniciu/Ramesh/Meer and for KL Tracker give a search with Birchfield + KLT

Best wishes,

Prithwi

[ferchuhill]

have you some information for

sharing, I am very concerned

[pguha77]

What kind of Information???

[megablablatruch]

it does not see the man standing still in the center

[abhikamune83]

really very good work done.congrates!!!!!!!!!

i am beginner in this domain pls suggest me some good paper for video tracking.

thank u

abhijit.kamune@gmail.com

[protikmaitra]

@ateendra123 I could be wrong but I am taking a guess:

Occlusion? Not handled.

Detection? At every frame.

How? Background subtraction and then region properties.

After detection, use the region to extract appearance template and then track.

Linking detected object between consecutive frames? Mean Shift or Kalman Filter.

I would let the creator correct me wherever I am wrong.

[ateendra123]

hi, how do you manage to handle occlusion? please let me know how do you manage to determine the size of the object?

[szproxy]

Is this real time? or is it post processing? which algorithm are u using?

[VIKINGONEGRO]

I am really interested on your research, My thesis is about recognizing peasants by their faces, but, part of the preprocessing is doing something similar to your work... Where can i get your article/paper? thanks in advance!!

[usamazahid1]

nice video want 2 learn ....how???