[vskiezv]

Hi fellow members on AI Dreams. I am looking at the feasibility of developing a program to extract site areas (eg. building areas) on google map with the given input of building addresses available. However, extracting area from google map is not so straightforward as the area has to be measured using the measuring tool before obtaining its area.
I have no background in programming and is unable to gauge the feasibility of this project. Hence, is this project possible using algorithmic coding or machine learning?

[LOCKSUIT]

Absolutely doable. You should use a efficient, multi-kill-1-bird technique instead of machine learning for this.

Go to the website called Upwork and post a hiring job, they can easily do it and even do some of the figuring out themselves too even without extra pay (note, Lock's programs are Lock's, but they do do the coding figuring out, i just narrate it haha). It's really easy. There's thousands of programmers on there :p. EVEN A CATAGORY FOR MACHINE LEARNING THEY R CALLING ME. ALSO ALL IMAGINABLE CATEGORIES KORR LIKE DATA EXTRACTION AND AND OMG MR. HINTON!!!! And freaking cheap in a sense just find a cheaper one but not too cheap, US computer scientists are at 70USD an hour while some do 5USD an hour from remote funny countries haha . You can get it done in 3 days from now, really. Don't learn programming, it takes years. Keep focused on the ideas, and hiring for implementation TESTING of your ideas, use all your time. The efficiency/RAM/memory/CPU etc can be studied as knowledge, so, you can stay in the idea zone while avoid programming. For example knowing how many bytes does a node take up? Ram is faster than SSD. GPU can be super fast for parallel. Knowing which python lists are less memory (ok, this gets into code I think...but hey, your AI idea is just that and matters, yes the efficiency matters just as much so that it is multi-generalizable/runnable, but that depends on both the network or system design plus the code specifics).

One thing that interested me just now and has before is types of datastructures, not sure even what are they haha, nets/trees?:
Primitive types.
Composite types or non-primitive type.
Abstract data types.
Arrays.
Lists.
Binary trees.
B-trees.
Heaps.
...interesting, time for more research!!

[Art]

Just a note:

The images you see via Google Earth/Maps can vary from 1 to 3 years old and is certainly not real time by any stretch.
What you see on Maps today as a blank field might now have a multi-story building in that location.

Good luck and don't be afraid to learn whatever you feel will best assist you in reaching your goals!

[WriterOfMinds]

I have no familiarity with the Google Maps API, so take this with a grain of salt, but ...

If you are just trying to input an address into Google Maps and capture a thumbnail image of the map area around the pointer, that sounds very straightforward. No machine learning needed, just geometry.

If you want to, say, identify all the buildings in a satellite image retrieved from Maps, that is something a neural net could probably help you with.  Face recognition exists now (though I gather it is not perfect) so I suspect you could do it.  How much work it would be, or whether someone has already trained an object-recognizer on building top-views specifically, I don't know.

[vskiezv]

Absolutely doable. You should use a efficient, multi-kill-1-bird technique instead of machine learning for this.

Go to the website called Upwork and post a hiring job, they can easily do it and even do some of the figuring out themselves too even without extra pay (note, Lock's programs are Lock's, but they do do the coding figuring out, i just narrate it haha). It's really easy. There's thousands of programmers on there :p. EVEN A CATAGORY FOR MACHINE LEARNING THEY R CALLING ME. ALSO ALL IMAGINABLE CATEGORIES KORR LIKE DATA EXTRACTION AND AND OMG MR. HINTON!!!! And freaking cheap in a sense just find a cheaper one but not too cheap, US computer scientists are at 70USD an hour while some do 5USD an hour from remote funny countries haha . You can get it done in 3 days from now, really. Don't learn programming, it takes years. Keep focused on the ideas, and hiring for implementation TESTING of your ideas, use all your time. The efficiency/RAM/memory/CPU etc can be studied as knowledge, so, you can stay in the idea zone while avoid programming. For example knowing how many bytes does a node take up? Ram is faster than SSD. GPU can be super fast for parallel. Knowing which python lists are less memory (ok, this gets into code I think...but hey, your AI idea is just that and matters, yes the efficiency matters just as much so that it is multi-generalizable/runnable, but that depends on both the network or system design plus the code specifics).

Thanks for the advice LOCKSUIT, that's a really interesting yet confusing read for me 
I'm actually an undergraduate, majoring in sustainable building engineering exploring this idea for my final year project. Which means that I have to be the one doing the coding required hahaha

If you are just trying to input an address into Google Maps and capture a thumbnail image of the map area around the pointer, that sounds very straightforward. No machine learning needed, just geometry.

Yes that's what I'm looking to do. I have a very long list of addresses (approximately 100K+) and is thinking of coming up with a program to search for these addresses, draw an area around the building and note down the area measured into Excel automatically.
However, with my very limited knowledge in programming (I've touched the surface of C++ before) and a duration of 6 months,

- Do I have enough time to complete this project?
- What are the most compatible languages to develop it?
- What are the required tools/knowledge (eg. Google API) to develop it?   

[WriterOfMinds]

Hmm, so ... it sounds like, after supplying an address and retrieving the corresponding image, you want to process it to figure out how big the building is, rather than just returning a cropped image of the same size every time.  That could be tricky and require some object recognition work.  You'll have to distinguish buildings from adjacent parking lots and roads.

It doesn't sound impossible, but with minimal programming experience and no knowledge of the specific tools involved, were I you I'd be hesitant to attempt it in six months.  (I'm assuming that the six months would not be spent exclusively working on this project; if your university is like mine you will be up to your knees in other classwork at the same time.)

Python is my go-to programming language to recommend for general use, and if you are interested in machine learning, I think there are libraries available for Python.  I don't know what interfaces best with the Google tools, though.

[goaty]

That would be cool,  its a bit like lawnmower man releasing your evil ai to live on the internet.

[Korrelan]

Hi Vskiezv

Google obviously doesn’t have the boundary locations of any plots.  That just leaves satellite photo analysis to find the boundaries, and I’m sure you’ll appreciate this task is difficult enough for a human, let alone a machine.

There is a huge range in both resolution and contrast between different areas on satellite view. There are also so many varied types of boarder, fence, hedge, road, stream, etc and they are also usually obscured by trees, buildings, etc. 



You could use an intelligent fill routine to find an open area (above), but look at the church, the perimeter is obscured by both trees and the building.  Object/ texture analysis could identify single trees and thus give the centre trunk locations. Straight line/ boarder analysis could predict straight obscure boarder runs, etc.

In my humble opinion, whilst possible because humans do it, this would be extremely difficult to automate using traditional image analysis and the narrow AI available in free libraries, especially in six months starting from scratch.

 

[HS]

Ah, so you too have met the dreaded C++... You might find this familiar. hehehe