Machine Vision - Motion Detection and Tracking

Objective: To build a vision system for detecting motion in an indoor environment, and for tracking the detected object based on the color distribution of the object. The input video feed is captured using a single, color-camera which is mounted on a steady non-mobile platform. Method[High level]: Use a background image of the scene [without the moving objects] as a reference frame. Do a pixel to pixel subtraction between the background image and every subsequent image frame, and this tells you where the motion is. Using the color information of the detected moving object, assign an id to it, and in the next frames look which of the detected object match closest to each of the objects detected and assigned an id, in the previous frame, assign the same id to the closest match, and save the color information of that object in the current frame, which would be used for matching objects in the next frames. Do the same for each of the detected object in each frame.

Category:

Science & Technology

Tags:

• computer vision
• machine vision
• vision
• motion detection
• tracking
• university of edinburgh
• bangalore
• advance vision
• image processing

[nash911]

.....in less than four days and for a negligible percentage of the total course marks.

Hope that answered your questions. Cheers!

[MrGargtarun84]

Hi nice work............

cn i have the VI for above application

i also want to implement it

Er. tarun garg

Lecturer BGIET Punjab india

[nash911]

@MrGargtarun84 Hi, what do you mean by 'VI'? Do you want the source code?

[nash911]

......object, calculate the color distribution, use Bhattacharyya method to classify it as one of the objects identified in the previous frame, or as a new object if the color difference is huge. And so you can see in the video, at time the previously green object becomes red, even though the red object, in the previous frame, was on the other side of the screen. Could have used a Kalman Filter with Condensation Tracking, to hypothesise the position of the moving object. But this was done......

[waqastro124]

So you are using only color information as an RGB vector? No further statistical manipulation? 2 questions:

1. Does this mean that ur algorithm works well only under conditions when moving objects have a markedly different color from the background?

2. How do u actually detect where the object is in the next frame? I mean do u use some kind of interpolation, least square error calcuation, etc. what?

I wud be interested to know, as Ive done sum work with my students on motion tracking. thanks.

[nash911]

....are completely different and +1 being perfectly identical.

1. Yes, the motion detection part of the algorithm works only when the moving object has a different color than the background, as i am doing a background subtraction. But the difference needn't have to be big, it works fine if for example the background is black and the moving object is dark grey in color.

2. No, i am not using any technique in particular for detecting the object in the next frame. I just detect the moving......

[waqastro124]

I know what Bhattacharya distance is.

Ok so this was for course, right. I was wondering that if it was a full-blown project, then it wasnt too much work, because simple frame subtraction etc. is not like too difficult. But great job in 4 days.

[nash911]

Hai, like mentioned in the description, its a very high level explanation of the method. Well, I am taking the RGB information of two objects [In consecutive frames] and calculating the distribution [That is calculating the number of pixels that has the value 0, 1,2.....255] for R,G and B separately, then use a method called Bhattacharyya Distribution [Google this up] on the color data, which tells me how likely the pair of objects have the same color distribution, where -1 indicates they......