finding dense location and define centeriod of the each dense location

I have attached .mat file and below zoomed figure of one section , I would like to find the centeriod of the dense location.

Since I have huge data around 400000 * 2, i am not able to use DBSCAN.

can you please do help me out with the loaded data ?

[darova]

Here is how your data looks like. What about dividing it into several parts?

[Adam Danz]

It looks like the y-values are clustered for some of the data. The diagonal line on the right is less obvious but I can still see some clustering. Perhaps you could explore some of Matlab's custer methods on the y-coordinates to determine where the separations occur. That's the bulk of the work. Once the data are grouped, you can measure each group's center of mass.

[SatyaPrakash Gupta]

I request you to zoom the figure and then visualize it , you can there are 2 paths.

Can you please do help me out to write the code for the data i have provided ?

[darova]

I saw your data zoomed. I know that it's the same as you show on th epicture before

If you can't use dbscan because of data size you can divide your data into several parts.

Did you try something else: clusterdata?

[SatyaPrakash Gupta]

No I have not tried clusterdata instead i used K-mean to divide the whole set of data into N clusters and further I have perform DBSCAN on each clusters and the result look good but still i miss of of the locations(90% i achieve the good value )

[darova]

Can i ask why do you need so many centroids? What are you doing further with them

[SatyaPrakash Gupta]

@darova :

Yes at the end i need to connect those line to each centriod where i am stuck now.

please find the attached data where i have found the centriods.

do you have any idea how can we connect them ?

or

how to further separate the data which below to left side and right side ?

Answer 1

[darova]

Here is simple but coarse (maybe not precise method)

Fill matrix Connect pixels Thin line Remove 'spur'

Results (green point - rounded orinal data)

I suppose you want to order those points now (connect them)

[SatyaPrakash Gupta]

Can you please do help me with the code to connect them , because i am not able to do it ?

it would really greatful to you if you could help me to connect them ?

[darova]

What about this trick?

And then just sort X coordinate

[SatyaPrakash Gupta]

Sorry this trick did not helped me, now i have a data separated with left and right , is it possible to connect a single line on left and single line on right side .

The line should break if the distance is more than 10m moving forward and start again with a new line for distance smaller than 10m

[darova]

I don't understand the question. This is your data:

Can you make a simple sketch of the result you expect?

Answer 2

[Image Analyst]

Looks like you already have an answer that you've accepted, however one way to approach it is to use fitPolynomialRANSAC() - it's in the Computer Vision Toolbox, if you have it. You can use this function twice. The first time it will get the first curve (one of them, not sure which one though). Then remove those points close to the first curve, and run it again to get the second curve and points close to it. fitPolynomialRANSAC() will return the polynomial coefficients of the curve and the indexes of the data points that are close to that curve.

Also see RANSAC in Wikipedia if you're not familiar with the RANSAC concept.

[SatyaPrakash Gupta]

sorry i do not have computer vision toolbox , how can i implement in matlab lab without this toolbox ?

[Image Analyst]

Well, don't bother. You already have an answer that you've accepted that works.

Answer 3

[SatyaPrakash Gupta]

yes , let me explain you in details.

load leftRightCoord.mat

scatter(leftRightCoord.left(:,1),leftRightCoord.left(:,2)) %% this will plot the xy coordinates appearing on the left side

hold on

scatter(leftRightCoord.right(:,1),leftRightCoord.right(:,2)) %% this will plot xy coordinates apearing on the right side

please find the figure for above explaination.

Now expected as below:

the right xy coordinates(blue color) should be coonected to next line asn similarly for left coordinates.

If the distance between xy coordinates are more than 10meters, then stop continue drawing the same line

else , continue drawing , like the below figure attached.

[darova]

Please use code button for code

There is also no need of such big images

[darova]

As for your question: what about my previous code? Use it and you can separate each lines using bwlabel

[SatyaPrakash Gupta]

I tried with your previous code but it did not help .

Can you help me with the code which fits my expectation

[darova]

I can't, sorry. Your question is too difficult for me

I did my best. That is all i can

[SatyaPrakash Gupta]

No problem, thank you very much for your support.

you have provided me the link for all the clustering methods , that has helped me a lot