1. Optical Flow

Let us start learning OpenCL with Optical Flow which will cover OpenCL APIs, Handling Mathematics in OpenCL, Memory Constrain and Efficiency.

Abstract:

Lets take up this as our first assignment and build a custom wrapper for OpenCL in C++, rather than using Khronos C++ wrapper.

Content:

• OpenCL Intro

• OpenCL Wrappers

• What is Optitcal Flow?

• Maths Involved

OpenCL Intro:

Let us create a new page here.

OpenCL Wrappers:

Why we need a wrapper? Answer is simple we are lazy and we don't want to waste our precious time in repeating the same stuff again and again, rather we can use that time for watching a movie or we can go for a dating or simply we can write some stupid things like this. So, we are going to use our grey matter a little and write some stupid code to cover cleverly written stupid code to do some ******* functionality.

We will take http://opencl.codeplex.com/SourceControl/latest as our reference code. Don't worry most of today's IT world is running with this little invention of Ctlr + C and

Ctlr + V. We start a new page here.

Design Considerations:

• Coding Standards used -> Come we will go to Coding section

• Qt Project Settings

• Abstract the Platform Initialization

What is Optical Flow?

Let us visit my old blog to get some data. Come with me.

References:

1. http://en.wikipedia.org/wiki/Optical_flow

The motion computed in the 2-D space is called Optical Flow.

Motion field = Real world 3D motion "How we see"

Optical flow field = Image intensity movement "How computers see"

In simple words, what we have to calculate is, a velocity vector (u,v) for every pixel, giving us the information how fast (u) and in which direction (v) that pixel is in motion.

Motion:

Movement of an object in the real world is in 3-D motion(XYZ), where as the same in the Image plane is only 2-D motion(XY). Significantly we have lost 1/3rd of the data, nonetheless we still can use these two co-ordinates motion information to do some very useful tasks.

Two Methods:

1. Lucas/Kanade

2. Horn/Schunck

Application:

• Games based on Phone movement

• Motion based Segmentation

• Structure from Motion

• Alignment - Camcorder video stablization

• - UAV Video Anlaysis

• Video Compression

Maths Involved:

Man Hours:

19/02/2014 - 2hours