Convert Time in timeseries object from posix to datetime

Hello,

I want to convert the Time in a timeseries object called GPSts (which currently is posixtime with nanoseconds e.g. "1604328037.75777") to datetime in the format 'HH:mm:ss.SSS' e.g. "23:46:11.4570".

I can convert the posixtime to the desired datetime using:

datetime(round(1000*GPSts.Time),'ConvertFrom','epochtime','Epoch', '1970-01-01', "TicksPerSecond",1000,'Format','HH:mm:ss.SSSS')

but when I try to modify the value of GPSts.Time I get the following errors:

Check for missing argument or incorrect argument data type in call to function 'isnan'.

Error in timeseries.tsChkTime (line 573)

if any(isinf(time)) || any(isnan(time))

Error in tsdata.timemetadata/reset (line 265)

t = timeseries.tsChkTime(t);

Error in timeseries/set.Time (line 154)

this.TimeInfo = reset(this.TimeInfo,input);

Please see my complete code example:

clc
clear

rosbagFolder = 'C:\Users\georg\OneDrive\Desktop\Rosbag';
filePattern = fullfile(rosbagFolder, '*.bag');
rosbagFiles = dir(filePattern);
GPSts = timeseries();
for i = 1 : length(rosbagFiles)
    bag = rosbag(rosbagFiles(i).name);
    bagSelGPS = select(bag, 'Time',...
        [bag.StartTime bag.EndTime], 'Topic', '/gx5/gps/fix');
    MsgsGPSTimeseries = timeseries(bagSelGPS, 'Latitude', 'Longitude', 'Altitude');
    GPSts = append(GPSts, MsgsGPSTimeseries);
    
end


GPSts.Time = datetime(round(1000*GPSts.Time),'ConvertFrom','epochtime','Epoch', '1970-01-01', ... 
    "TicksPerSecond",1000,'Format','HH:mm:ss.SSSS')  %Error occurs here
gpsFrameDuration = median(diff(GPSts.Time));
gpsFrameDuration.Format = "s";
gpsRate = 1/seconds(gpsFrameDuration);

[Sylvain]

Your code works fine, so it seems you import NaN data, flagged as error.

[Georg Novotny]

Hello Sylvian,

thank you for your reply!

I don't think that

datetime(round(1000*GPSts.Time),'ConvertFrom','epochtime','Epoch', '1970-01-01', ... 
    "TicksPerSecond",1000,'Format','HH:mm:ss.SSSS')

returns NaN.

Please see the output of said command:

ans =

1067×1 datetime array

14:40:37.7580

14:40:38.0080

14:40:38.2490

14:40:38.5020

14:40:38.7510

14:40:38.9980

14:40:39.2510

14:40:39.5090

14:40:39.7480

Answer 1

[Georg Novotny]

Here my working code:

clear
close all

Read rosbag and store data in variables
rosbagFolder = 'C:\Users\georg\OneDrive\Desktop\Rosbag';
filePattern = fullfile(rosbagFolder, '*.bag');
rosbagFiles = dir(filePattern);
allMsgsGPS = cell(0,0);
allMsgsLIDAR = cell(0,0);
for i = 1 : length(rosbagFiles)
    bag = rosbag(rosbagFiles(i).name);
    bagSelGPS = select(bag, 'Time',...
        [bag.StartTime bag.EndTime], 'Topic', '/gx5/gps/fix');
    MsgsGPS= readMessages(bagSelGPS, 'DataFormat','struct');    
    allMsgsGPS =[allMsgsGPS; MsgsGPS];
end

Convert data to matlab Mat format
gpsTime = cellfun(@(t) datetime( double(t.Header.Stamp.Sec)+double(t.Header.Stamp.Nsec)*10^-9, ...
    'ConvertFrom','posixtime', 'Format', 'HH:mm:ss.SSSS'), allMsgsGPS);
gpsLatitude = cellfun(@(m) double(m.Latitude), allMsgsGPS);
gpsLongitude = cellfun(@(m) double(m.Longitude), allMsgsGPS);
gpsAltitude = cellfun(@(m) double(m.Altitude), allMsgsGPS);

gpsSequence = timetable(gpsTime, gpsLatitude, gpsLongitude, gpsAltitude);

gpsFrameDuration = mean(diff(gpsSequence.Time));
gpsFrameDuration.Format = "s";
gpsRate = 1/seconds(gpsFrameDuration);
fprintf('GPS  : %s, %3.1f Hz\n', char(gpsFrameDuration), gpsRate);

Answer 2

[Peter Perkins]

Georg, three suggestions:

1) Numeric precision may be better if you separate the conversion of whole and fractional seconds like this:

datetime(double(t.Header.Stamp.Sec),  'ConvertFrom','posixtime', 'Format', 'HH:mm:ss.SSSS') + milliseconds(+double(t.Header.Stamp.Nsec)*10^-6)

2) Setting the datetime format to 'HH:mm:ss.SSSS' suggests that maybe you could be using durations, not datetimes. If what you want is "time since midnight", you can get that with the timeofday function.

3) I'm not actually very familiar with ROS or the ROS Toolbox, and if your code works, that's great, but if allMsgsGPS is a cell array of scalar structs all with the same fields, each contianing a scalar string, you might consider something like the following, which is likely to be more performant than what you have:

>> s1.a = "1"; s1.b = "4";
>> s2.a = "2"; s2.b = "5";
>> s3.a = "3"; s3.b = "6";
>> c = {s1;s2;s3}
c =
  3×1 cell array
    {1×1 struct}
    {1×1 struct}
    {1×1 struct}
>> s = vertcat(c{:})
s = 
  3×1 struct array with fields:
    a
    b
>> a = double([s.a])';
>> b = double([s.b])';
>> t = table(a,b)
t =
  3×2 table
    a    b
    _    _
    1    4
    2    5
    3    6

[SL: added missing ' in the datetime call.]