多传感器信息融合.pptVIP

1　多传感器信息融合 2 Kalman filtering Contents Background Example of Kalman filtering Arithmetic Application Recommended reading Rudolf Emil Kalman 匈牙利数学家 BSMS at MIT PhD at Columbia 1960年发表的论文《A New Approach to Linear Filtering and Prediction Problems》 Use for 机器人导航、控制 传感器数据融合 雷达系统以及导弹追踪 计算机图像处理 头脸识别 图像分割 图像边缘检测 Evolution Kalman filter Extended Kalman filter (EKF) Unscented Kalman filter (UKF) Particle fliter (PF) Sequential importance sampling (SIS) Unscented particle filter (UPF) Auxiliary sampling importance resampling filter (ASIR) Regularized particle filter (RPF) Gaussian particle filter (GPF) Compare——KF A Linear Dynamical System is a partially observed stochastic process with linear dynamics and linear observations, both subject to Gaussian noise. Compare——EKF Based on local linearization of KF Easy to realize and has good performance in Gaussian and mild nonlinear environment. Compare——UKF Utilizes a set of definite samplings to approximate posterior probability density function Suitable for any non-linear environment but Gaussian environment Compare——PF Utilizes a set of random particles to approximate posterior probability density function Plays good act in any non-linear and non-Gaussian environment Compare——PF UPF: use UKF Arithmetic to update ASIR: import the other one important function RPF: for dealing with new problems bring by resampling GPF: use Gaussian distributing approximate posterior distributing, avoid degeneracy problem Contents background Example of Kalman filtering Arithmetic Application Recommended reading Temperature Problem - Real World 假设当前室内温度仅跟上一时刻有关 但变化中可能有噪声（White Gaussian Noise ） 温度计观测（摄氏－〉华氏） 读数会有误差（White Gaussian Noise ） 两种噪声相互无关 根据连续的观测值来推算实际温度变化 Temperature Problem - Ideal World First: For: 你相信温度是恒定的，所以你会得到k时刻的温度预测值是跟k-1时刻一样的，假设是23度，同时该值的高斯噪声的偏差是5度. Temperature Problem - Ideal World Afterward:温度计测得k时刻的温度值是25度，同时该值的偏差是4度。 For:估算k时刻的实际温度有两个温度值:23度和25度。 Use covariance to judge: So：实际温度值是23+0.78