MySQL 8.0

By using db<>fiddle, you agree to license everything you submit by Creative Commons CC0.
create table YourLatiLongiTudesTable
(
  ID int primary key auto_increment, 
  lat decimal(8,5) not null, 
  lng decimal(8,5) not null,
  `timestamp` timestamp not null
);
insert into YourLatiLongiTudesTable
(lat, lng, `timestamp`) values                              
('23.21465', '84.37856', '2016-01-01 00:00'),
('23.21465', '84.37856', '2016-01-01 00:03'),
('23.21584', '84.37877', '2016-01-01 00:06'),
('23.21584', '84.37877', '2016-01-01 00:09'),
('23.21465', '84.37856', '2016-01-01 00:12'),
('23.21465', '84.37856', '2016-01-01 00:15'),
('23.21465', '84.37856', '2016-01-01 00:18'),
('23.21465', '84.37856', '2016-01-01 00:21')
;
--
-- Using variables
-- First calculate a rank on descending date
-- Then calculate a rownum based on the rank
--
SELECT id, lat, lng, `timestamp`
-- , rnk, rnk_rn
FROM
(
  SELECT *
  , CASE 
    WHEN rnk != @prev_rnk
     AND @prev_rnk := rnk
    THEN @rn := 1
    WHEN @prev_rnk := rnk
    THEN @rn := @rn + 1
    END AS rnk_rn
  FROM
  (
    SELECT t.*
    , CASE 
      WHEN lat != @prev_lat
       AND @prev_lat := lat 
      THEN @rnk := @rnk + 1
      WHEN @prev_lat := lat 
      THEN @rnk
      END AS rnk
    FROM YourLatiLongiTudesTable t
    CROSS JOIN (SELECT @rnk:=0, @prev_lat:=0) v
    ORDER BY `timestamp` desc, id desc
  ) q1
  CROSS JOIN (SELECT @rn:=0, @prev_rnk := 0) v
  ORDER BY `timestamp`, id
) q2
WHERE rnk = 1 
  AND rnk_rn = 1
;


id
lat
lng
timestamp


5
23.21465
84.37856
2016-01-01 00:12:00


--
-- Using window functions
-- First calculate a rank by summing changes.
-- Then calculate a rownum based on the rank
--
SELECT id, lat, lng, `timestamp`
-- , rnk, rnk_rn
FROM
(
  SELECT *
  , ROW_NUMBER() OVER (PARTITION BY date(`timestamp`), rnk ORDER BY `timestamp`, id) AS rnk_rn
  FROM 
  (
    SELECT *
    , SUM(IF(lat=prev_lat,0,1)) OVER (PARTITION BY date(`timestamp`) ORDER BY `timestamp` DESC, id DESC) AS rnk
      FROM 
      (
        SELECT *
        , LAG(lat) OVER (PARTITION BY date(`timestamp`) ORDER BY `timestamp` DESC, id DESC) AS prev_lat
        FROM YourLatiLongiTudesTable t
      ) q1
   ) q2
) q3
WHERE rnk = 1
  AND rnk_rn = 1
ORDER BY `timestamp`, id;


id
lat
lng
timestamp


5
23.21465
84.37856
2016-01-01 00:12:00