I am back in Pittsburgh after a short trip to Portland (better known as the Garden of Eden). I went back and saw (some) friends and visited the old high school. Everybody seems done with the whole college phase of their life and ready to move on to the you're-gonna-do-this-until-75-or-whenever-the-retirement-age-is phase or the feeling-self-pity-since-life-has-dicked-you-over-until-retirement-age phase. Little do we know that by the time we reach retirement, the government will have succeeded in passing the mandatory death act of 2060 for unproductive members of society.

CHS is as decadent as ever. Word has it that they had 20 inductees to the academic wall last year, the vast majority of which never took humanities...which also happens to be a shadow of its former Andrew Holt Browning self...I remember when we spent many an afternoon in the cafeteria seeing if we could scrap together enough points to get on the wall...many deserving individuals did not make it...now it seems that packing your pockets with AP classes is the easiest way to have a plaque in your honor.   

Its snowing again. I recieved word while I was in Portland that it was almost 70 in Pittsburgh and in fact fairly temperate from dec 22nd to jan 4, the exact dates of my departure. Portland of course was at its worst during my stay...never reaching above 58 and raining every day...there was no brown snow of course, but thats because there is a god...My friend brought up that I must have some power over the seasons, much like Persephone. Except that my aura of bitterness brings plague, famine and cold wherever I go.  

Cincinnati has the worst airport in the world. Never fly America West.

Grad Apps - check
GRE - check
Appearance Modeling - check
Subterranean - check
Agrobotics - check
Compilers - check
NASA project - check
NASA tech. report - check
TAing 15-212 - oh my god freaking check!

That felt so good - I hadn't slept in 2 days. Probably the toughest semester I've ever had is finally over with. I just finished my demo for the volumetric terrain scanner today - it went well - and I'm off to the airport soon. See you all in Oregon.

Next semester will be all parties and sleep, I swear.

Currently Listening Labyrinth der Sinne see related

So I finally got my laptop back ... and it works... for now. Bestbuy has been spared the wrath of the killbots for now.

Its official, I am now a TA for 15-212. Finally, I get to help teach a class in the field I love, semantics (not ECE!). Everyone must now sign up for it that hasn't yet. Thats an order!

So I got my laptop back from "BestBuy" on wednesday. Notice how putting "BestBuy" in quotes denotes it is a steaming pile of human refuse. I originally sent it in to get my busted video card fixed before my 3 year warantee thingy expired. The funniest thing is when they take a month to fix your computer, it comes back, and they aint fixed jack. Oh wait, I missed the joke, was that supposed to be funny? So I took it back to "BestBuy," and this ridiculous hoodlum starts poking various places on my laptop as if his mysterious powers of exorcism will fix my burned out riser card...all the while I could swear he was talking about how the "flux capacitors" in toshiba laptops were suboptimal. Moron. I just waited to say something like "I've been building computers since you were suckling your mother's teat: 2 or 3 years, asshole." So now I have to wait another 10 business days so that they can actually do their job. If they haven't fixed it then, i'm setting the killbots loose. My degree in hyperkinetic mortality automata will finally mean something.

Currently Listening The Hand That Feeds By Nine Inch Nails see related

So heres a little about what i've been doing. I've been hard at work devising method to extract stable landmarks in an environment from pictures. What this translates to is recording "memorable" objects that are associated with a location and can later be recalled to estimate if a location has been visited. This is much harder than anything I've ever done, and I've had many years experience in computer vision.

This leads to an important clarification. Vision is easy, cognition is not. Producing a static 2d representation of an environment is nothing more than excitations of cones and rods in the eye, each one forming a "picture element," or pixel if you will . Technically, retinex theory suggests that the level of excitation of each individual rod or cone is not important, but rather the brain considers the excitation of each element in comparison to all other elements, a sort of normalized matrix of color values. This partially explains the human robustness to shadows and ambient lighting. Even our shadow detection and texture recognition ability is nothing more than a few tedious linear transformations on a color grid. An operation using no more than a fraction of second on a fast computer.

Yet, meaningful computer vision continues to elude us. We can detect shadows, match scenes, and even reconstruct partial images using a computer. But we cannot do what humans do simply. Robots cannot perceive because mathematics does not infer. It can tell you that a tree in a picture taken from a certain angle is the same as that in another picture, but it cannot tell you that a tree is indeed a "tree." That is why roboticists in computer vision rely on probabilities and patterns. Everyone knows that in an outdoor scene, the probability of blue skies, white clouds and green trees is high. Which is why we want to find objects of interest that are none of those. Yet, the probabilty of an object being truly unique is proportional to the probability that it is a phantom, an illusion created from a peculiar combination of lighting or physical effects. And since an object is nothing more than a combination of pixels there is nothing more than primitive edge or outline detection to tell if an object in our world of bits really corresponds to a physically independent shape. This is bad unless there was a way to infer what a unique object in vision might mean in the physical world and hence extract the probability that a feature is emphemeral.

This is where I come in. I am hoping there is some sort of pattern to stable keypoints. That is, certain regions of interest that persist if the sun moves, the wind blows or the camera is tilted a bit. I'm betting that a combination of these stable points will correspond to physical features that we can comprehend. If there is a way to quantify stability, uniqueness should be easy. Or maybe not. Right now we're throwing everything in the book at it. W're removing shadows, analyzing entropy and looking at textures. Theres still a long summer ahead...

Currently Listening Find You're Here By Wolfsheim see related