Machine Vision: A Pre-Emptive

[Xittenn]

I'm planning on doing some work with machine vision. I'm not looking at anything to complex like the AI personal assistants and such, or to simple like for industrial and manufacturing purposes(by simple I mean cost effective!) Thinking on the lines of like an atom with a ten gig static ram drive(maybe more but size is important) and Aptina high res image sensors; so cost effective! Looking at simply doing the exact opposite of what's being done with the current animation studios and gaming engines in terms of softcoding. Reading shade gradients, extrapolating the lighting. Observing stencil shading, defining object space. Reading shading gradients, defining object, modeling objects spacial coordinates as mesh. Observe movement sequence, assign nodal points. Assess data table, interpret, allocate object label in conjunction with pre-programmed objects list. So really simple stuff. I don't even know what GPU to look into 'cause I know doing the opposite of what they do is not something they will probably do well...................... Just thought maybe yall might have some input for me before I start!

[Pangloss]

Not really, but it sounds like fun. I'd be interested in seeing anything you'd care to share with us.

[Mokele]

There are several textbooks on the subject that should be on amazon and/or google books. IME, MATLAB has an excellent image processing toolbox that can be programmed very efficiently.

 

Mokele

[Xittenn]

MATLAB, hmmmmmmmmmmmmmm, don't see anything that suggests the availability of an include for C++. Never used a math programming language before. Probably should; they are most obviously much better at it than I. I'm going to program my own math library, I like the control and require the practice. Hmph I don't even understand Lie Groups......................I've got to stop napping so much; so little time so much to do.

[Mokele]

It's actually called MATLAB because it's all matrix math and operations, which sounds weird and takes a lot of getting used to, but is quite powerful. When I took a machine vision course, we used it exclusively, and in my current lab, we use a special program developed for it for image analysis.

 

Back when I took the machine vision course, I had one bit of input for the class as the sole biologist, and I think it may have helped - don't shoot for human-level. Humans actually have *amazingly* good day vision for animals, only outdone by birds. Most animals do quite well with far less resolution and less sophisticated processing. So basically, don't overshoot.

 

Mokele

[Xittenn]

I was shooting for reasonable recognition of spheres and cubes(maybe a couch) to start..................................human vision that's funny....................resolution 1024*768..............wow biologist, machine vision course I'm jealous.....................OP, OP.............regardless I'm still pretty bent on programming my own math library, it's a learning experience! Was thinking of going really substructure like and programming the whole thing in machine code but that's being silly.

---

Merged post follows:

Consecutive posts merged

I'm not just going for machine vision it's just my primer. What ever happened to fuzzy logic systems. It was pretty big in the nineties but I haven't heard anything in the last decade. I should do a search, one of those recurrent thoughts that will soon happen. Suggestions..................... I've been tempted to start the project from the hardware side first. Not just selecting the components but actually designing and manufacturing them myself. I have a really long list of things needing to be built just for these purposes of which I will avoid................uv laser circuit board etcher, a mechanical hand/arm like the one in Ironman to machine parts for me on hyper-poly axis in lathe and mill fashions, a chem chamber, kiln, fuel cell to power it all(I don't watch TV, I don't have a TV, except on holidays at my Dads; stop laughing.) At any rate current products are pretty much benign tumors growing off of the ears of the masses. I really want to do something new and cool and I don't have a fancy degree to get me into Intel. I do have time and some money and I do have a good deal of experience and self education. So if anybody has any cool random thoughts on the topic to throw at me.................Mokele I would love to hear what you have to say on hardware possibly some good use of array technology to make brain chips I would love to hear them. I know I said Atom but I really meant..........................

---

Merged post follows:

Consecutive posts merged

With respect to Machine Vision...........................

---

Merged post follows:

Consecutive posts merged

Well this is really important to me and getting some of you wonderful people into a conversation about it is a task I would like to accomplish. So I'll just keep posting snippets about what I want to do and how I propose to achieve it until I accomplish my goal.....................this could take a while!

 

A good qualitative overview of what I hope to accomplish is:

 

"Robotic vision which can monitor a room for activities and recognize the people, activities and objects within!"

 

So pretty tough stuff but that's no excuse for not trying.

 

SPATIAL RECOGNITION

 

As I pointed out earlier on in this thread I want to base my spatial recognition on shading gradients, light source and the more obvious quantitative observations of shadows(the normal stuff.) I cannot at this time see any means of using colour to measure space especially at the near proximity intended. I'm sure there may be some use in a human brain for colour and space with respect to diffusion of light around objects. I assume these effects are minute and for my purposes insignificant if not unreadable within a two meter radius. Not to mention a processing guzzling task.

 

It would probably be a good idea to develop a database of objects with a known given size. This would be useful assuming there is the capability of recognizing the object in the first place.

 

So I guess a good place to start would be in assessing the resolution of the shading gradient required to read good spatial depth. I should probably also look deeper into stereoscopic vision and how shading plays a role within it.

 

Something I've personally never heard anyone reference to that I think may play a role is observable to those who can take off their glasses. Have you ever noticed how an unfocused image will produce a spiraling copy of the image. eg. looking at the moon without glasses produces many moons spiraling out from the center moon.

 

anywho not writing a book here...........................

---

Merged post follows:

Consecutive posts merged

Well I gave some good thought on the whole shading gradient resolution thing. I've decided that rather than subjecting monkeys to spatial recognition tests I will just go with standard 24bit true colour and remove as necessary from there. It doesn't really make a difference to the architecture and it will save me making upgrades as I go. I thought of going Fairchild and limiting to grey scale but that was just being well.......................

 

Regardless I do believe I will be taking the first steps into this project within the next few weeks. I have no intention of doing an online project overview but hope to maintain this thread hopefully exchanging a variety of ideas on this topic with others.

Edited January 31, 2009 by buttacup
Consecutive posts merged.

[Mokele]

IMHO, you'll find much more success focusing on stereoscopic vision than on shading for spatial perception. A lot of animals have vision that's too poor to really resolve shading, but can easily figure out distances via having 2 (or more) eyes.

[Xittenn]

No argument here. I'm just looking at greater detail of resolution as opposed to simpler despite necessity; although greater is necessary as facial recognition is required(alternatives.) When I said spatial recognition I meant distance from and depth within. I was also leaving the topic open to all possible methods of sight analysis. I was seriously suggesting new or generally ignored methods of analysis eg. of light diffraction around objects, light and nodal interference(the multi moon effect, I don't know how to properly describe this one.) Regardless just brainstorming sometimes the redundant and the ridiculous provide a better means to an ends or even put light on a new solution.

 

Mokele any suggestions on reading material with regard to neural physics and the visual cortex with heavy indications on array technology and tissue engineering......................

Edited February 1, 2009 by buttacup

[Mokele]

Two things come to mind:

 

1) Many animals which are excellent at sight don't simply have a uniform array of photoreceptors, but also have a cluster in the center of the eye that has super-high-resolution, called the fovea. When humans see, we constantly move our eyes around, focusing the fovea on different points, and the brain patches it together into single high resolution image. Of course, wiggling the eyes that much would require a complex mechanical setup.

 

2) In most animals, there's a 'neighbor' effect. The output of a photoreceptor is more intense if those next to it are receiving a different color, which has the result of emphasizing edges. This actually happens at the level of retina, so all images your brain receives essentially go through this simple edge-enhancing program before they get there.

[Xittenn]

Two things come to mind:

 

1) Many animals which are excellent at sight don't simply have a uniform array of photoreceptors, but also have a cluster in the center of the eye that has super-high-resolution, called the fovea. When humans see, we constantly move our eyes around, focusing the fovea on different points, and the brain patches it together into single high resolution image. Of course, wiggling the eyes that much would require a complex mechanical setup.

 

2) In most animals, there's a 'neighbor' effect. The output of a photoreceptor is more intense if those next to it are receiving a different color, which has the result of emphasizing edges. This actually happens at the level of retina, so all images your brain receives essentially go through this simple edge-enhancing program before they get there.

 

1) more minute..................how would you best describe the blur caused by a seeing impairment; myopia. Radial blur was what I would have assumed but.............. I'm saying blur isn't radial in nature but a series of less intense copies of an image created by light interference patterns and that this is possibly useful for sight analysis..............the picture is always blurred but when it's view focused the pictures are all in the right spot. I don't believe the two ideas are the same, correct me if I'm wrong.

 

2) Although I was not referring to something someone would truly be seeing, it would be a good use of minute colour variations caused by light diffraction near the edges of any object. I guess I do see edges a little more exaggerated than they truly are.

 

Thanks for the feedback!

 

No good reads:embarass:!

[Mokele]

how would you best describe the blur caused by a seeing impairment; myopia. Radial blur was what I would have assumed but.............. I'm saying blur isn't radial in nature but a series of less intense copies of an image created by light interference patterns and that this is possibly useful for sight analysis..............the picture is always blurred but when it's view focused the pictures are all in the right spot. I don't believe the two ideas are the same, correct me if I'm wrong.

 

Myopia is simply improper focusing, resulting in a generally blurry image without sharp lines or contrast. I don't know enough about interference patterns to answer the other part.

 

Mokele

[Xittenn]

The blur part is the point of interest radial or interference patterns...Radial could be ignored, interference patterns could have impact on the way we see stuff!

---

Merged post follows:

Consecutive posts merged

Well regardless of observations on blur which no one else but me and my optometrist seem to share..................

 

Optix

So like I'm thinking gel optics. It in itself could be a study. I'm sure this one is under current development with machine vision and/or medical developments of replacement eyes...................Books, Journals, Societies.................Anyway I think it would be cool to develop something around gel optics in an attempt to simulate a human eye and all the mechanicals.

---

Merged post follows:

Consecutive posts merged

http://www.wipo.int/pctdb/en/wo.jsp?IA=GB1994001421&DISPLAY=DESC

 

well after all of five seconds of looking I found this..............at least I won't be treading on anyones patents.

 

reason number 5million and 1 I should finally by my own computer..............remember kermit, so do I...............remember owning a copmuter, ummmmmmmmmm.

 

Anyway, in terms of chemistry I'm looking at a long list of substances contained within Canadas DSL. It's so hard being freelance and maintaining good insight into the legality when you are not attempting to do anything outside of the law. Most of the equipment is readily purchased and/or fabricated by moi. Implimentation is my biggest concern. Aside from what I'm seeing as the most obvious difficulties...........proper dimensioning, actually finding a material that will envelope the lens, deform the lens and be itself easily fused to a motor control system; not to mention the need for regular hydration...........this sounds like fun. I wonder how well an aerogel lens would work with a cmos Image sensor..................

 

Should this be in engineering????????

Edited February 4, 2009 by buttacup
Consecutive posts merged.

[Mokele]

I wouldn't bother trying to recreate a biological eye. Remember, evolution doesn't produce an optimum, but rather simply a cobbled-together organ that's merely sufficient for the animal's needs. Hell, the human retina is backwards.

[Xittenn]

http://darwin-online.org.uk/content/frameset?itemID=F381&viewtype=side&pageseq=1 page 205

 

"suppose that the eye, with all its inimitable contrivances for adjusting the focus to different distances, for admitting different amounts of light, and for the correction of spherical and chromatic aberration, could have been formed by natural selection, seems, I freely confess, absurd in the highest possible degree. When it was first said that the sun stood still and the world turned round, the common sense of mankind declared the doctrine false; but the old saying of Vox populi, vox Dei, as every philosopher knows, can never be trusted in science. Reason tells me, that if numerous gradations from a perfect and complex eye to one very imperfect and simple, each grade being useful to its possessor, can be shown to exist; if further, the eye does vary ever so slightly, and the variations be inherited, which is certainly the case; and if any variation or modification in the organ be ever useful to an animal under changing conditions of life, then the difficulty of believing that a perfect and complex eye could be formed by natural selection, though insuperable by our imagination, can hardly be considered real. How a nerve comes to be sensitive to light, hardly concerns us more than how life itself first originated; but I may remark that several facts make me suspect that nerves sensitive to touch may be rendered sensitive to light, and likewise to those coarser vibrations of the air which produce sound."

 

-Charles Darwin

 

How time does change things.........................

---

Merged post follows:

Consecutive posts merged

so that being said...................yah!

 

Sight the Next Generation

Well I personally had no intention on improving sight but hey.........."no better time than here, no better time than now............."

 

What would you do to improve human sight ........ would it be advantageous to make replacement eyes that could prove far superior to our own ............ would they look good, look natural ........... how would they feel?

 

I could go for some binoc/telescopic vision and add some microscopy. Infrared? Capable of emitting fluorescense, laser .......... Raman Spectroscopy for all ................. I'm pretty sure the dolls I plan on making don't need this but hey, it's a broad study they can just have the ability no one needs to know. Better refresh rate ............ suggestions .............. higher resolution ......... I like the idea of using gel optics for the natural appearance I'm not sure how far I can stretch them to achieve some of these improvements.................. 60 second warning damn!

---

Merged post follows:

Consecutive posts merged

I wonder how much Hyaluronic Acid would be needed to maintain hydration while emitting laser light.................say 5 watts............hmmmmmmmmmm this could get expensive!

---

Merged post follows:

Consecutive posts merged

New thought.....as much fun as Raman Spectroscopy is........

 

"Hi Bob!" Says Sally

"Well hello there Sally." Replies Bob

"Come on in."

"Hey I hear those things do Raman Spectroscopy.........."

"Are you talking about my doll Bob?"

"Yes Sally."

"That's !@#$^% up."

 

At any rate. A small thin film LCD inset in the lens could be useful in projecting diagnostics...............I like this very much. It could be as simple as a hex editor or as complex as a low to high res desktop. I know this isn't really a standard component to machine vision but most applications therein could integrate it to great benefit.

---

Merged post follows:

Consecutive posts merged

soooooooo back to

 

vision analysis

ummmmmmmmmm yah!

 

=>Contrast image break it into 'Skin' groups =>

=>Read Gradients within groups, apply curve fitting, define 'B-Splines'=>

=>Redefine 'B-Splines' as defined 'Lie Groups'=>

=>Define 'Global Objects' from sets of defined 'Lie Groups'=>

=>Define 'Object' from 'Global Objects'=>

=>Assemble 'Spline Model'=>

=>Convert to 'Mesh Model' (possibly Subdivision Surfaces, I like Mesh)=>

 

Whether the objects will be compared as spline models or mesh models, well........ How much analysis should be done in 2D and/or 3D, well....... how far I'm off on this generalization, well.............

 

I wish I could go over some code for Maya and maybe Tier 0 and Tier 1 for the Source Engine but yeah.............. A lot of code is being patented I should probably go over some of that......... various texturing stuffs and modeling techniques........ I've read a book on graphics coding. I haven't read anything on Machine Vision. I'm kind of afraid to, I'm sure it will be informative but the focus will undoubtedly be out of context............ at the moment! Not that animation and gaming engines aren't out of context, they are after all backwards. I just think this approach will serve me better in that it's more fitting to broad analysis in real time as opposed to focused analysis which a good deal of machine vision applications are at this time.

Edited February 8, 2009 by buttacup
Consecutive posts merged.