Linear interpolation in rows of matrix

Hi!

I have vehicle speed measurement data in the following form (small example):

Distance (from the point of measurement, in metres): 1, 2, 3, 4, 5, 6, 7

Speed of vehicle1 at these points (in km/h): 50, 54, NaN, 62, 62, 64, NaN

Speed of vehicle2 at the points: 40, NaN, NaN, NaN, 48, 54, 60

Speed of vehicle3 at the points: NaN, 60, 62, 62, 66, NaN, NaN

Speed of vehicle 4 at the points: 44, 47, NaN, 55, NaN, NaN, 58

...

I would like to get a value instead of NaN with linear interpolation. BUT I dont want to interpolate if NaNs are the first/last elements of the measurement.

Needed result of example above: [50 54 58 62 62 64 NaN; 40 42 44 46 48 54 60; NaN 60 62 62 66 NaN NaN, 44 47 51 55 56 57 58].

Im absolutely a novice user of MATLAB, but I can not handle this problem in Excel, so I really need your help. How to insert these data (in a matrix form or how?), and how to get a good result with that I can work further (there are quite a lot vehicles in this database :S) ?

Thank you in advance

Gábor

[Mohammad Sami]

If your data is in a table. use the function fillmissing.

Use the option 'EndValues','none' to avoid filling in the first and the last values.

Answer 1

[Star Strider]

The fillmissing function (R2016b and later) will do what you want:

D = [1, 2, 3, 4, 5, 6, 7];
V1 = [50, 54, NaN,  62, 62, 64, NaN];
V2 = [40, NaN, NaN, NaN, 48, 54, 6];
V3 = [NaN, 60, 62, 62, 66, NaN, NaN];
V4 = [44, 47, NaN, 55, NaN, NaN, 58];

Vs = [V1;V2;V3;V4];

Vout = fillmissing(Vs,'linear',2, 'EndValues','none')

producing:

Vout =
    50    54    58    62    62    64   NaN
    40    42    44    46    48    54     6
   NaN    60    62    62    66   NaN   NaN
    44    47    51    55    56    57    58

As requested.

[Star Strider]

Gabor Pauer’s Answer moved here —

Thank you very much Star Strider, it is just perfect :)

One more question:

Is there a quick solution in MATLAB to solve the same problem, if the initial data is not so well structured?

Here is the above example in the form i have (first column is the distance from measurement points, 2nd column is the speed there). I can structure this data in the above mentioned form in excel, but that also takes a lot of time. Maybe can someone solve it in a qick way in MATLAB?

THANKS :)

Dist Speed

1 50

2 54

4 62

5 62

6 64;

1 40

5 48

6 54

7 60;

2 60

3 62

4 62

5 66;

1 44

2 47

4 55

7 58;

[Star Strider]

My pleasure.

Tha appears to be inconsistent, since the distances recycle.

This will interpolate or extrapolate the second column of ‘DS’ to the distances in ‘dist’:

DS = [1        50
      2        54
      4        62
      5        62
      6        64;
      1        40
      5        48
      6        54
      7        60;
      2        60
      3        62
      4        62
      5        66;
      1        44
      2        47
      4        55
      7        58;];

rd = [1; find(diff([0; DS(:,1)]) <0); size(DS,1)];                              % New Start Indices
dist = 1:7;                                                                     % Interpolation/Extrapolation Distance Vector
for k = 1:numel(rd)-1
    idxrng = rd(k):rd(k+1)-1;
    DSI(:,k) = interp1(DS(idxrng,1), DS(idxrng,2), dist, 'linear','extrap');
end

producing the interpolated or extrapolated values for the second column (the first column is always the ‘dist’ vector):

DSI =
    50    40    58    44
    54    42    60    47
    58    44    62    51
    62    46    62    55
    62    48    66    59
    64    54    70    63
    66    60    74    67

[Gabor Pauer]

Wow Star Strider, many thanks again, I tried it on my data and it works :)

Just one more thing: this solution does not take into account, that if the first or last value(s) are missing, then leave it empty (or put there a NaN), which was solved in the first method with the EndValues command.

(e.g. in my example we dont have measurement data for vehicle1 at the last measurement point, where distance is 7.

So First column of DSI should be: 50,54,58,62,62,64,NaN

Third column should be: NaN, 60, 62, 62,66, NaN, NaN )

Can you maybe solve this easily? I don't want to waste your time, many thanks friend once again!

[Star Strider]

As always, my pleasure!

There were no NaN values in those data. If there were, the best approach would likely be to use the same essential logic to separate the segments, and then use fillmissing instead of interp1 on each segment.