预测
预测线性卡尔曼滤波器的状态估计误差协方差
Syntax
描述
Syntaxes for Predefined Motion Model
Use these syntaxes if you specify a predefined motion model in theMotionModel属性trackingKF目的。
无需控制输入的自定义运动模型的语法
如果指定了这些语法MotionModelproperty as"Custom"和do not use control inputs.
语法与控制Inpu定制运动模型t
如果指定了这些语法MotionModelproperty as"Custom"并使用控制输入。
例子
Create Constant-Velocity Linear Kalman Filter
Create a linear Kalman filter that uses a 2D constant velocity motion model. Assume that the measurement consists of thexy-对象的位置。
Specify the initial state estimate to have zero velocity.
x = 5.3;y = 3.6;初始状态= [x; 0; y; 0];kf = trackingkf（'MotionModel','2D Constant Velocity','State',initialState);
在恒定轨迹上为对象创建测量位置。
VX = 0.2;vy = 0.1;t = 0.5;pos = [0：vx*t：2;5：vy*t：6]';
Predict and correct the state of the object.
fork = 1:size(pos,1) pstates(k,:) = predict(KF,T); cstates(k,:) = correct(KF,pos(k,:));end
Plot the tracks.
plot(pos(:,1),pos(:,2),“ K。”,pstates(:,1),pstates(:,3),“+”,。。。cstates(:,1),cstates(:,3),“ O”）xlabel（"x [m]"）ylabel（"y [m]") grid xt = [x-2, pos(1,1)+0.1, pos(end,1)+0.1]; yt = [y, pos(1,2), pos(end,2)]; text(xt,yt,["First measurement",“第一个位置”,"Last position"]) legend("Object position",“预测位置”,“校正位置”)
Use CustomtrackingKFwith Control Inputs
指定一个10秒的模拟时间，时间为1秒。
rng(2021)% For repeatable results仿真时间= 20;dt = 1;tspan = 0：dt：仿真时间;步骤=长度（tspan）;
Specify the motion model as a 2-D constant velocity model with a state of [x;vx;y;vy]。测量是[x;y]。
A1D = [1 dt; 0 1]; A = kron(eye(2),A1D)％状态转运顿模型
A =4×41 1 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 1
h1d = [1 0];h = kron（眼睛（2），H1D）％测量模型
H =2×41 0 0 0 0 0 1 0
sigma = 0.2; R = sigma^2*eye(2);％测量噪声协方差
Specify a control model matrix.
B1D = [0; 1]; B = kron(eye(2),B1D)％控制模型矩阵
b =4×20 0 1 0 0 0 0 1
Assume the control inputs are sinusoidal on the velocity components,vx和vy。
gain = 5; Ux = gain*sin(tspan(2:end)); Uy = gain*cos(tspan(2:end)); U =[Ux; Uy];％控制输入
Assuming the true initial state is [1 1 1 -1], simulate the system to obtain true states and measurements.
初始状态= [1 1 1 -1]';％[m m/s m m/s]truestates = nan（4，步骤）;truestates（：：，1）=初始状态;fori=2:steps trueStates(:,i) = A*trueStates(:,i-1) + B*U(:,i-1);endmeasurements = H*trueStates + chol(R)*randn(2,steps);
可视化真实的轨迹和测量。
figure plot(trueStates(1,:),trueStates(3,:),"DisplayName",“真相”） 抓住上plot(measurements(1,:),measurements(2,:),"x","DisplayName","Measurements"）xlabel（“ x（m）”）ylabel（“ y（m）”) legend
Create atrackingKF使用自定义运动模型过滤。通过指定控制模型来启用控件输入。根据第一个测量指定过滤器中的初始状态。
初始FilterState = [测量（1,1）;0;测量（2,1）;0];filter = trackingkf（“ MotionModel”,"Custom",。。。"StateTransitionModel"，一个，。。。“ MeasurementModel”,H,。。。“控制模型”，b，。。。"State"，初始FilterState）;
通过使用预测和correct对象功能。
estimateStates = NaN(4,steps); estimateStates(:,1) = initialFilterState;fori = 2:steps predict(filter,U(:,i-1)); estimateStates(:,i) = correct(filter,measurements(:,i));end
Visualize the state estimates.
绘图（估算attates（1，:)，估算attates（3，:)，，，，"g","DisplayName","Estimates"）；
输入参数
Output Arguments
Extended Capabilities
Version History
See Also
