-
Notifications
You must be signed in to change notification settings - Fork 4
/
pqvicrUKF.m
264 lines (205 loc) · 8.39 KB
/
pqvicrUKF.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
% Position, orientation and velocity UKF
%clc
close all
%% UKF parameters
ukf_alpha = .01;
ukf_beta = 2;
%% x: state vector
% Composed by IMU position, rotation quaternion ([w x y z]') and velocity
% in the world frame
% [p_w_i; q_w_i; v_w_i] = x(1:10);
%% P: state covariance matrix 9-by-9
% NOTE: P is not 10-by-10 because we consider 3 deg of freedom for q_w_i
%% u: process inputs
% Composed by measured IMU acceleration and rotational velocity
% u = [accel_i_measured(1:3, i); gyro_i_measured(1:3, i)]
%% n: process noise
%% Q: process noise covariance matrix
Qacc = eye(3) * 0.1^2;
Qrot = eye(3) * 0.001 * 300 * pi/180; % rad/s
Q = [Qacc zeros(3); zeros(3) Qrot];
%% z: measurements
% See section 4.3 Measurement Model on page 11
% z is a 2n-by-1 column vector of observed pixel coordinates in the
% form [x1 y1 ... xn yn]' where n is the number of 3D feature points
%% R: measurement noise covariance matrix
% The associated block-daigonal covariance matrix of z
% R = diag(R1 ... Rn) = 0.1^2 * eye(length(z));
%% Starting index
i = 2;
j = 2;
nowTime = imuData(i-1,3);
%% Initial estimate
clear x;
expected_rad_error = 1 * pi / 180;
init_rad_error = 0.2* expected_rad_error;
err_quat = matrix2quaternion(rotx(init_rad_error)*roty(init_rad_error)*rotz(init_rad_error));
x = [p_w(:,i); q_w_i(:,i); v_w(:,i-1); p_i_c+.1*randn(3,1); quaternionproduct(err_quat,q_i_c)]; % easy as ground truth location
%x = [p_w(:,i); q_w_i(:,i); v_w(:,i-1); p_i_c+.1*randn(3,1); q_i_c]; % easy as ground truth location
xstart=x;
Ppos = diag([.5 .5 .5]);
Pori = (10 * pi / 180)* eye(3);
Pvel = diag([0.5 0.5 0.5]);
Ppic=diag([.5 .5 .5]);
Pqic=(20*pi/180)*eye(3);
%P = [Ppos zeros(3, 6); zeros(3) Pori zeros(3); zeros(3,6) Pvel];
P=blkdiag(Ppos,Pori,Pvel,Ppic,Pqic);
Pstart=P;
%% Initialize storage matrices and figure
numCamMeasurements = size(observed_pts_c, 2);
numImuMeasurements = length(imuData);
numPoses = numImuMeasurements + numCamMeasurements;
accumPoses = zeros(3,numPoses);
accumQuat = NaN * ones(4,numPoses);
distanceError = zeros(1, numPoses);
picError=zeros(1,numPoses);
qicError=zeros(1,numPoses);
velocityError = zeros(1, numPoses);
orientationError = zeros(1, numPoses);
process_params = cell(4,1);
obs_params = cell(5,1);
h = figure('Name','Position, Orientation and Velocity Estimation', ...
'NumberTitle','off','Position',[10 10 1000 600]);
%% Begin Kalman filter
count = 1;
while (i <= numImuMeasurements && j <= numCamMeasurements )
% Read the timestamp for the next data input
imuTime = imuData(i,3);
camTime = camData(j,3);
% Get previous orientation belief
prev_q = x(4:7);
prev_qic_q=x(14:17);
if (imuTime <= camTime)
%% Prediction step
pastTime = nowTime;
nowTime = imuTime;
dt = nowTime - pastTime;
u = [accel_i_measured(:,i); gyro_i_measured(:,i)];
process_params{1} = u;
process_params{2} = dt;
process_params{3} = prev_q;
process_params{4} = gravity;
process_handle = @processModelPQVICr;
%x_se = [x(1:3); 0; 0; 0; x(8:10)]; % State error vector with q in MRP
x_se = [x(1:3); 0; 0; 0; x(8:10);x(11:13); 0;0;0];
%[x_se, Ppred] = predictUKF(x_se, process_handle, process_params, ...
% P(1:end-3,1:end-3), Q, ukf_alpha, ukf_beta);
[x_se, Ppred] = predictUKF(x_se, process_handle, process_params, ...
P, Q, ukf_alpha, ukf_beta);
mrp_error = x_se(4:6);
% Convert MRP update vector to quaternion update
norm_mrp_error = sqrt(sum(mrp_error.^2, 1));
dq0 = (1 - norm_mrp_error) ./ (1 + norm_mrp_error);
q_error = [ dq0;
bsxfun(@times,(1+dq0),mrp_error)];
quat_new = quaternionproduct(q_error, prev_q);
quat_new = quat_new./norm(quat_new);
mrp_qic_error=x_se(13:15);
norm_mrp_qic_error=sqrt(sum(mrp_qic_error.^2,1));
dq_qic0=(1-norm_mrp_qic_error)./(1+norm_mrp_qic_error);
q_qic_error=[dq_qic0; bsxfun(@times,1+dq_qic0,mrp_qic_error)];
quat_qic_new=quaternionproduct(q_qic_error, prev_qic_q);
quat_qic_new=quat_qic_new/norm(quat_qic_new);
x'
%x = [x_se(1:3); quat_new; x_se(7:9);x(end-2:end)];
x = [x_se(1:3); quat_new; x_se(7:9);x_se(10:12);quat_qic_new];
x'
%P=blkdiag(Ppred,P(end-2:end,end-2:end));
P=Ppred;
i = i + 1;
else
%% Correction Step
% Perform correction step
z = noisy_observed_pts_c(:,j);
R = 0.1^2 * eye(length(z));
x_se = [x(1:3); 0; 0; 0; x(8:10);x(11:13);0;0;0]; % State error vector with q in MRP
ukf_N = length(x_se);
%p_IMU_camera = repmat(p_i_c, 1, 2*ukf_N+1);
%q_IMU_camera = repmat(q_i_c, 1, 2*ukf_N+1);
p_world_pts = pts_w(1:3, :);
obs_params{1} = prev_q;
obs_params{2} = prev_qic_q;
obs_params{3} = p_world_pts;
obs_params{4} = K;
obs_handle = @measurementModelPQVICr;
[ x_se, P ] = correctUKF( x_se, P, R, z, obs_handle, obs_params, ukf_alpha, ukf_beta );
mrp_error = x_se(4:6);
% Convert MRP update vector to quaternion update
norm_mrp_error = sqrt(sum(mrp_error.^2, 1));
dq0 = (1 - norm_mrp_error) ./ (1 + norm_mrp_error);
q_error = [ dq0;
bsxfun(@times,(1+dq0),mrp_error)];
quat_new = quaternionproduct(q_error, prev_q);
quat_new = quat_new./norm(quat_new);
mrp_qic_error=x_se(13:15);
norm_mrp_qic_error=sqrt(sum(mrp_qic_error.^2,1));
dq_qic0=(1-norm_mrp_qic_error)./(1+norm_mrp_qic_error);
q_qic_error=[dq_qic0; bsxfun(@times,1+dq_qic0,mrp_qic_error)];
quat_qic_new=quaternionproduct(q_qic_error, prev_qic_q);
quat_qic_new=quat_qic_new/norm(quat_qic_new);
x = [x_se(1:3); quat_new; x_se(7:9);x_se(10:12);quat_qic_new];
j = j + 1;
end
if (i < numImuMeasurements)
%% Distance error
distanceError(1,count) = norm(x(1:3) - p_w(:,i-1));
velocityError(1,count) = norm(x(8:10) - v_w(:,i-1));
orientationError(1,count) = findQuaternionError(x(4:7), q_w_i(:,i-1));
picError(1,count)=norm(x(11:13)-p_i_c);
qicError(1,count)=findQuaternionError(x(14:17),q_i_c);
%% Plot
accumPoses(:,count) = x(1:3);
count = count + 1;
if mod(count, 10) == 1
figure(h);
clf
subplot(5,2,[1, 3, 5, 7, 9]);
plot3(accumPoses(1,1:count-1), accumPoses(2,1:count-1), accumPoses(3,1:count-1),'-');
hold on;
plot3(p_w(1,1:i), p_w(2,1:i), p_w(3,1:i), 'g');
% hold on;
% plot3(pts_w(1, :), pts_w(2, :), pts_w(3, :), 'r.');
axis equal
axis vis3d
xlabel('x'); ylabel('y'); zlabel('z');
grid on;
title('Motion Estimation');
subplot(5,2,2);
plot(1:count,distanceError(1:count));
maxErr = max(distanceError);
axis([0 numPoses 0 maxErr]);
% xlabel('Time');
ylabel('Squared Error');
title('Distance Error');
subplot(5,2,4);
plot(1:count,velocityError(1:count));
maxErr = max(velocityError);
axis([0 numPoses 0 maxErr]);
% xlabel('Time');
ylabel('Squared Error');
title('Velocity Error');
subplot(5,2,6);
plot(1:count,orientationError(1:count));
maxErr = max(orientationError);
axis([0 numPoses 0 maxErr]);
xlabel('Time');
ylabel('Squared Error');
title('Orientation Error');
subplot(5,2,8);
plot(1:count,picError(1:count));
maxErr = max(picError)+.1;
axis([0 numPoses 0 maxErr]);
xlabel('Time');
ylabel('Squared Error');
title('IMU-Camera Translation Error');
subplot(5,2,10);
plot(1:count,qicError(1:count));
maxErr = max(qicError)+.1;
axis([0 numPoses 0 maxErr]);
xlabel('Time');
ylabel('Squared Error');
title('IMU-Camera Rotation Error');
%pause
end
end
end