4 * Copyright (c) 2014 Samsung Electronics Co., Ltd.
6 * Licensed under the Apache License, Version 2.0 (the "License");
7 * you may not use this file except in compliance with the License.
8 * You may obtain a copy of the License at
10 * http://www.apache.org/licenses/LICENSE-2.0
12 * Unless required by applicable law or agreed to in writing, software
13 * distributed under the License is distributed on an "AS IS" BASIS,
14 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
15 * See the License for the specific language governing permissions and
16 * limitations under the License.
21 #ifdef _ORIENTATION_FILTER_H_
23 #include "orientation_filter.h"
25 //Windowing is used for buffering of previous samples for statistical analysis
26 #define MOVING_AVERAGE_WINDOW_LENGTH 20
28 #define GRAVITY 9.80665
30 //Needed for non-zero initialization for statistical analysis
31 #define NON_ZERO_VAL 0.1
32 //microseconds to seconds
33 #define US2S (1.0 / 1000000.0)
34 //Initialize sampling interval to 100000microseconds
35 #define SAMPLE_INTV 100000
36 #define ACCEL_THRESHOLD 0.2
37 #define GYRO_THRESHOLD (0.01 * PI)
39 // constants for computation of covariance and transition matrices
40 #define ZIGMA_W (0.05 * DEG2RAD)
42 #define QWB_CONST ((2 * (ZIGMA_W * ZIGMA_W)) / TAU_W)
43 #define F_CONST (-1 / TAU_W)
44 #define SQUARE(T) (T * T)
46 #define NEGLIGIBLE_VAL 0.0000001
48 #define ABS(val) (((val) < 0) ? -(val) : (val))
51 template <typename TYPE>
52 orientation_filter<TYPE>::orientation_filter()
54 TYPE arr[MOVING_AVERAGE_WINDOW_LENGTH];
56 std::fill_n(arr, MOVING_AVERAGE_WINDOW_LENGTH, NON_ZERO_VAL);
58 vect<TYPE, MOVING_AVERAGE_WINDOW_LENGTH> vec(arr);
67 m_magnetic_alignment_factor = 1;
69 m_gyro.m_time_stamp = 0;
72 template <typename TYPE>
73 orientation_filter<TYPE>::~orientation_filter()
77 template <typename TYPE>
78 inline void orientation_filter<TYPE>::initialize_sensor_data(const sensor_data<TYPE> *accel,
79 const sensor_data<TYPE> *gyro, const sensor_data<TYPE> *magnetic)
81 m_accel.m_data = accel->m_data;
82 m_accel.m_time_stamp = accel->m_time_stamp;
86 unsigned long long sample_interval_gyro = SAMPLE_INTV;
88 if (m_gyro.m_time_stamp != 0 && gyro->m_time_stamp != 0)
89 sample_interval_gyro = gyro->m_time_stamp - m_gyro.m_time_stamp;
91 m_gyro_dt = sample_interval_gyro * US2S;
92 m_gyro.m_time_stamp = gyro->m_time_stamp;
94 m_gyro.m_data = gyro->m_data * (TYPE) PI;
96 m_gyro.m_data = m_gyro.m_data - m_bias_correction;
99 if (magnetic != NULL) {
100 m_magnetic.m_data = magnetic->m_data;
101 m_magnetic.m_time_stamp = magnetic->m_time_stamp;
105 template <typename TYPE>
106 inline void orientation_filter<TYPE>::orientation_triad_algorithm()
108 TYPE arr_acc_e[V1x3S] = {0.0, 0.0, 1.0};
109 TYPE arr_mag_e[V1x3S] = {0.0, (TYPE) m_magnetic_alignment_factor, 0.0};
111 vect<TYPE, V1x3S> acc_e(arr_acc_e);
112 vect<TYPE, V1x3S> mag_e(arr_mag_e);
114 vect<TYPE, SENSOR_DATA_SIZE> acc_b_x_mag_b = cross(m_accel.m_data, m_magnetic.m_data);
115 vect<TYPE, V1x3S> acc_e_x_mag_e = cross(acc_e, mag_e);
117 vect<TYPE, SENSOR_DATA_SIZE> cross1 = cross(acc_b_x_mag_b, m_accel.m_data);
118 vect<TYPE, V1x3S> cross2 = cross(acc_e_x_mag_e, acc_e);
120 matrix<TYPE, M3X3R, M3X3C> mat_b;
121 matrix<TYPE, M3X3R, M3X3C> mat_e;
123 for(int i = 0; i < M3X3R; i++)
125 mat_b.m_mat[i][0] = m_accel.m_data.m_vec[i];
126 mat_b.m_mat[i][1] = acc_b_x_mag_b.m_vec[i];
127 mat_b.m_mat[i][2] = cross1.m_vec[i];
128 mat_e.m_mat[i][0] = acc_e.m_vec[i];
129 mat_e.m_mat[i][1] = acc_e_x_mag_e.m_vec[i];
130 mat_e.m_mat[i][2] = cross2.m_vec[i];
133 matrix<TYPE, M3X3R, M3X3C> mat_b_t = tran(mat_b);
134 rotation_matrix<TYPE> rot_mat(mat_e * mat_b_t);
136 m_quat_aid = rot_mat2quat(rot_mat);
139 template <typename TYPE>
140 inline void orientation_filter<TYPE>::compute_accel_orientation()
142 TYPE arr_acc_e[V1x3S] = {0.0, 0.0, 1.0};
144 vect<TYPE, V1x3S> acc_e(arr_acc_e);
146 m_quat_aid = sensor_data2quat(m_accel, acc_e);
149 template <typename TYPE>
150 inline void orientation_filter<TYPE>::compute_covariance()
152 TYPE var_gyr_x, var_gyr_y, var_gyr_z;
153 TYPE var_roll, var_pitch, var_azimuth;
154 quaternion<TYPE> quat_diff, quat_error;
156 if(!is_initialized(m_quat_driv.m_quat))
157 m_quat_driv = m_quat_aid;
159 quaternion<TYPE> quat_rot_inc(0, m_gyro.m_data.m_vec[0], m_gyro.m_data.m_vec[1],
160 m_gyro.m_data.m_vec[2]);
162 quat_diff = (m_quat_driv * quat_rot_inc) * (TYPE) 0.5;
164 m_quat_driv = m_quat_driv + (quat_diff * (TYPE) m_gyro_dt * (TYPE) PI);
165 m_quat_driv.quat_normalize();
167 m_quat_output = phase_correction(m_quat_driv, m_quat_aid);
169 m_orientation = quat2euler(m_quat_output);
171 quat_error = m_quat_aid * m_quat_driv;
173 m_euler_error = (quat2euler(quat_error)).m_ang;
175 m_gyro.m_data = m_gyro.m_data - m_euler_error.m_ang;
177 m_euler_error.m_ang = m_euler_error.m_ang / (TYPE) PI;
179 m_gyro_bias = m_euler_error.m_ang * (TYPE) PI;
181 insert_end(m_var_gyr_x, m_gyro.m_data.m_vec[0]);
182 insert_end(m_var_gyr_y, m_gyro.m_data.m_vec[1]);
183 insert_end(m_var_gyr_z, m_gyro.m_data.m_vec[2]);
184 insert_end(m_var_roll, m_orientation.m_ang.m_vec[0]);
185 insert_end(m_var_pitch, m_orientation.m_ang.m_vec[1]);
186 insert_end(m_var_azimuth, m_orientation.m_ang.m_vec[2]);
188 var_gyr_x = var(m_var_gyr_x);
189 var_gyr_y = var(m_var_gyr_y);
190 var_gyr_z = var(m_var_gyr_z);
191 var_roll = var(m_var_roll);
192 var_pitch = var(m_var_pitch);
193 var_azimuth = var(m_var_azimuth);
195 m_driv_cov.m_mat[0][0] = var_gyr_x;
196 m_driv_cov.m_mat[1][1] = var_gyr_y;
197 m_driv_cov.m_mat[2][2] = var_gyr_z;
198 m_driv_cov.m_mat[3][3] = (TYPE) QWB_CONST;
199 m_driv_cov.m_mat[4][4] = (TYPE) QWB_CONST;
200 m_driv_cov.m_mat[5][5] = (TYPE) QWB_CONST;
202 m_aid_cov.m_mat[0][0] = var_roll;
203 m_aid_cov.m_mat[1][1] = var_pitch;
204 m_aid_cov.m_mat[2][2] = var_azimuth;
207 template <typename TYPE>
208 inline void orientation_filter<TYPE>::time_update()
210 euler_angles<TYPE> orientation;
212 m_tran_mat.m_mat[0][1] = m_gyro.m_data.m_vec[2];
213 m_tran_mat.m_mat[0][2] = -m_gyro.m_data.m_vec[1];
214 m_tran_mat.m_mat[1][0] = -m_gyro.m_data.m_vec[2];
215 m_tran_mat.m_mat[1][2] = m_gyro.m_data.m_vec[0];
216 m_tran_mat.m_mat[2][0] = m_gyro.m_data.m_vec[1];
217 m_tran_mat.m_mat[2][1] = -m_gyro.m_data.m_vec[0];
218 m_tran_mat.m_mat[3][3] = (TYPE) F_CONST;
219 m_tran_mat.m_mat[4][4] = (TYPE) F_CONST;
220 m_tran_mat.m_mat[5][5] = (TYPE) F_CONST;
222 m_measure_mat.m_mat[0][0] = 1;
223 m_measure_mat.m_mat[1][1] = 1;
224 m_measure_mat.m_mat[2][2] = 1;
226 if (is_initialized(m_state_old))
227 m_state_new = transpose(m_tran_mat * transpose(m_state_old));
229 m_pred_cov = (m_tran_mat * m_pred_cov * tran(m_tran_mat)) + m_driv_cov;
231 for (int j=0; j<M6X6C; ++j) {
232 for (int i=0; i<M6X6R; ++i) {
233 if (ABS(m_pred_cov.m_mat[i][j]) < NEGLIGIBLE_VAL)
234 m_pred_cov.m_mat[i][j] = NEGLIGIBLE_VAL;
236 if (ABS(m_state_new.m_vec[j]) < NEGLIGIBLE_VAL)
237 m_state_new.m_vec[j] = NEGLIGIBLE_VAL;
240 m_quat_9axis = m_quat_output;
241 m_quat_gaming_rv = m_quat_9axis;
243 m_rot_matrix = quat2rot_mat(m_quat_driv);
245 quaternion<TYPE> quat_eu_er(1, m_euler_error.m_ang.m_vec[0], m_euler_error.m_ang.m_vec[1],
246 m_euler_error.m_ang.m_vec[2]);
248 m_quat_driv = (m_quat_driv * quat_eu_er) * (TYPE) PI;
249 m_quat_driv.quat_normalize();
251 if (is_initialized(m_state_new))
253 m_state_error.m_vec[0] = m_euler_error.m_ang.m_vec[0];
254 m_state_error.m_vec[1] = m_euler_error.m_ang.m_vec[1];
255 m_state_error.m_vec[2] = m_euler_error.m_ang.m_vec[2];
256 m_state_error.m_vec[3] = m_state_new.m_vec[3];
257 m_state_error.m_vec[4] = m_state_new.m_vec[4];
258 m_state_error.m_vec[5] = m_state_new.m_vec[5];
262 template <typename TYPE>
263 inline void orientation_filter<TYPE>::time_update_gaming_rv()
265 euler_angles<TYPE> orientation;
266 euler_angles<TYPE> euler_aid;
267 euler_angles<TYPE> euler_driv;
269 m_tran_mat.m_mat[0][1] = m_gyro.m_data.m_vec[2];
270 m_tran_mat.m_mat[0][2] = -m_gyro.m_data.m_vec[1];
271 m_tran_mat.m_mat[1][0] = -m_gyro.m_data.m_vec[2];
272 m_tran_mat.m_mat[1][2] = m_gyro.m_data.m_vec[0];
273 m_tran_mat.m_mat[2][0] = m_gyro.m_data.m_vec[1];
274 m_tran_mat.m_mat[2][1] = -m_gyro.m_data.m_vec[0];
275 m_tran_mat.m_mat[3][3] = (TYPE) F_CONST;
276 m_tran_mat.m_mat[4][4] = (TYPE) F_CONST;
277 m_tran_mat.m_mat[5][5] = (TYPE) F_CONST;
279 m_measure_mat.m_mat[0][0] = 1;
280 m_measure_mat.m_mat[1][1] = 1;
281 m_measure_mat.m_mat[2][2] = 1;
283 if (is_initialized(m_state_old))
284 m_state_new = transpose(m_tran_mat * transpose(m_state_old));
286 m_pred_cov = (m_tran_mat * m_pred_cov * tran(m_tran_mat)) + m_driv_cov;
288 euler_aid = quat2euler(m_quat_aid);
289 euler_driv = quat2euler(m_quat_output);
291 if ((SQUARE(m_accel.m_data.m_vec[1]) < ACCEL_THRESHOLD) && (SQUARE(m_gyro.m_data.m_vec[0]) < GYRO_THRESHOLD))
293 if ((SQUARE(m_accel.m_data.m_vec[0]) < ACCEL_THRESHOLD) && (SQUARE(m_gyro.m_data.m_vec[1]) < GYRO_THRESHOLD))
295 if (SQUARE(m_gyro.m_data.m_vec[2]) < GYRO_THRESHOLD)
297 euler_angles<TYPE> euler_gaming_rv(euler_aid.m_ang.m_vec[0], euler_aid.m_ang.m_vec[1],
298 euler_driv.m_ang.m_vec[2]);
299 m_quat_gaming_rv = euler2quat(euler_gaming_rv);
304 if (is_initialized(m_state_new)) {
305 m_state_error.m_vec[0] = m_euler_error.m_ang.m_vec[0];
306 m_state_error.m_vec[1] = m_euler_error.m_ang.m_vec[1];
307 m_state_error.m_vec[2] = m_euler_error.m_ang.m_vec[2];
308 m_state_error.m_vec[3] = m_state_new.m_vec[3];
309 m_state_error.m_vec[4] = m_state_new.m_vec[4];
310 m_state_error.m_vec[5] = m_state_new.m_vec[5];
314 template <typename TYPE>
315 inline void orientation_filter<TYPE>::measurement_update()
317 matrix<TYPE, M6X6R, M6X6C> gain;
318 matrix<TYPE, M6X6R, M6X6C> iden;
319 iden.m_mat[0][0] = iden.m_mat[1][1] = iden.m_mat[2][2] = 1;
320 iden.m_mat[3][3] = iden.m_mat[4][4] = iden.m_mat[5][5] = 1;
322 for (int j=0; j<M6X6C; ++j) {
323 for (int i=0; i<M6X6R; ++i) {
324 gain.m_mat[i][j] = m_pred_cov.m_mat[j][i] / (m_pred_cov.m_mat[j][j] + m_aid_cov.m_mat[j][j]);
325 m_state_new.m_vec[i] = m_state_new.m_vec[i] + gain.m_mat[i][j] * m_state_error.m_vec[j];
328 matrix<TYPE, M6X6R, M6X6C> temp = iden;
330 for (int i=0; i<M6X6R; ++i)
331 temp.m_mat[i][j] = iden.m_mat[i][j] - (gain.m_mat[i][j] * m_measure_mat.m_mat[j][i]);
333 m_pred_cov = temp * m_pred_cov;
336 for (int j=0; j<M6X6C; ++j) {
337 for (int i=0; i<M6X6R; ++i) {
338 if (ABS(m_pred_cov.m_mat[i][j]) < NEGLIGIBLE_VAL)
339 m_pred_cov.m_mat[i][j] = NEGLIGIBLE_VAL;
343 m_state_old = m_state_new;
345 TYPE arr_bias[V1x3S] = {m_state_new.m_vec[3], m_state_new.m_vec[4], m_state_new.m_vec[5]};
346 vect<TYPE, V1x3S> vec(arr_bias);
348 m_bias_correction = vec;
350 m_gyro_bias = m_gyro_bias + vec;
353 template <typename TYPE>
354 void orientation_filter<TYPE>::get_device_orientation(const sensor_data<TYPE> *accel,
355 const sensor_data<TYPE> *gyro, const sensor_data<TYPE> *magnetic)
357 initialize_sensor_data(accel, gyro, magnetic);
359 if (gyro != NULL && magnetic != NULL) {
361 orientation_triad_algorithm();
363 compute_covariance();
367 measurement_update();
369 m_quaternion = m_quat_9axis;
371 } else if (!gyro && !magnetic) {
373 compute_accel_orientation();
375 m_quaternion = m_quat_aid;
379 orientation_triad_algorithm();
381 m_quaternion = m_quat_aid;
383 } else if (!magnetic) {
385 compute_accel_orientation();
387 compute_covariance();
389 time_update_gaming_rv();
391 measurement_update();
393 m_quaternion = m_quat_gaming_rv;
397 #endif //_ORIENTATION_FILTER_H_