**Start of term!**

Well… it’s that time of year again when a new chapter begins… The usual start-of-year stuff took place of course… Director of Studies Meeting, Tutor Meeting … (I’ve got a new tutor this term, Prof Osborn, while Dr Kusukawa is on sabbatical for the Michaelmas and Lent… )

It’s been quite quiet, but here are the highlights of the week…

**Lectures start**

Yep… still have those … although the privileges of being a 4th year mean that I only have to take 4 courses this term (and 4 next term…).

I’ve settled on, what I think are, some very interesting ones:

Though these seem perhaps a strange combination, there is a reason… I’m very keen on the first two (in the light of graduating with ‘Electrical Engineering’), I find control system design intuitive and generally useful, and the Innovation Management module covers almost exactly what I have been working on at the European Patent Office for the last 3 years.

I also decided on attending the Statistical Pattern Processing course out of interest and because I believe it may be useful material to use in my project (see below) when it comes to implementing a classification system. It’s full of *delightful* probability (posteriors, priors, likelihood), machine learning concepts, weird maths symbols and scary Gaussians though, so I expect it to be quite tough…

I also decided against taking the Digital Filters course… The course content is insanely useful in many areas and usually intuitive, but I just don’t think I’m wired to answer such questions in 20 minutes in an exam room in April… I’m intrigued however by the Kalman filter… so I might go back when they discuss that…

For those interested, find this term’s department timetable here

**Microsoft-related **

Not much this week, been busy with start of term, but I did have a phone discussion with the Live Mesh team in Seattle on Tuesday about … Live Mesh… More on this at PDC!

I also found a minute to throw together a Photosynth of Trinity College (click here), which I may use for an upcoming MSP competition…

**Fresher’s events**

I’m not a big ‘Fresher’s events’ person, but I did make it to the Fresher’s Fair on Tuesday, where most Cambridge societies have stalls set up trying to attract first year’s. I was mainly interested in seeing what my friend Hugo’s stall was like for the Cambridge University Wireless Society… of course he is trying to convince me to take my Foundation Radio License… maybe… and this of course led me to help out to carry big boxes of heavy radio equipment back to Corpus College… also in order to admire their new (scary!) clock, which I will be sure to grab a picture of this week…

I also gave led a Trinity Fresher’s tour of the Engineering Department on Tuesday, which was coupled with the usual questions of ‘Is there wi-fi in the department’, ‘What if we are late to lectures’, ‘What is Linux’ (most departmental computers run a version of Suse. It’s a nice system, but I still find it clumsy at times … and it only froze three times this week on me…)? This year’s freshers were very very nice and seem like a terrific group. They almost fainted when they saw the ‘Example Papers’ (Cambridge Problem Sheets) and the amount of work they would have to do in the first few weeks… 🙂 Lucky them… I remember the joys of Mechanics, Structures… ah the ‘good’ old days…

**Progress on 4th Year Project**

For those not familiar with my project, supervised by Ognjen Arandjelovic, this is it:

“To implement a state-of-the-art computer vision algorithm for the recognition of objects in video sequences”

Not trivial… I spent the week revising basic Maths concepts e.g. 2D convolution and processing of images in Matlab (my new best friend), read a few scientific papers, went to a project safety lecture, panicked at the amount of probability that there probably (get it?) will be in my project…

Essentially how I see it is (very simplistically put):

Sample images from the video –> extract interesting image features –> match these features to a feature ‘database’ –> recognize the object in the image.

I decided to try this week to implement my own version of a famous ‘image feature extraction’ algorithm by David Lowe called ‘SIFT’ – Scale Invariant Feature Transform. The algorithm is complicated, so I won’t detail it, but for those interested, I got as far as forming the pyramid of Difference of Gaussians (boy was the positioning in the Matlab figure hard… probably harder than the algorithm itself)…

When I implemented the first part of the keypoint localization routine (maxima/minima extrema detection in scale-space) my matlab program returned what I think are strange keypoints (shown below for the first level of the Difference of Gaussians Pyramid)… and it’s quite slow!

I’ve set SIFT aside for now and I’ll speak about it with my supervisor in our Tuesday meeting…

Yesterday, I continued the research, into topics such as Support Vector Machines, trying to understand Bayes’ Theorem for the 15th time (yuk…), ‘Histograms of Oriented Gradients’ as an alternative to SIFT, and I’m reading through a paper of ‘Pictorial Structures for Object recognition’…

Next week will be shaped by the first project meeting on Tuesday, where I’m hoping to clarify certain aspects of the implementation that I will be putting in practice…

I’m not sure into how much detail about my project I should go in this blog… let me know if you want to know more!

That’s it for this week!

Fascinating post. Thanks for posting

By:

MikiwJey8on May 11, 2010at 6:23 pm