Extended Karman Filter

Zhenglei 2018 January

This is a project to estimate the car position from Lidar data and Radar data. Lidar data is using Kalman filter. Radar data is using Extended Karman filter, because the radar data is provided by non-linear data(rho, rho-dot, phi). Then we put the two types of sensor data together, and get an more accurate estimate. The solution is so called sensor fusion.

The following gif picture is record from the simulator.

  • Blue circle: radar measurement
  • Red circle: laser measurement
  • Green circle : EKF estimated position

1.Create the enviroment for carnd-term2

I don't have an mac, and my working PC is install windows7

So, this is the options

  1. windows7 professinal + vmware(ubuntu), not work
  2. windows7 professinal + docker, not work
  3. pure ubuntu, it works

    ... I haven't test

So, this is the suggestion, if you are using windows7. Don't try anything on it. It will waste your time.

Buy a computer or borrow a computer, then install ubuntu (16.4 is the version use)

2. From a empty computer to run the code with Simulator

2.1 install ubuntu

   (PreRequirment: you already have another windows PC)

   on the windows pc:

   1)down ubuntu from https://www.ubuntu.com/download/desktop

   2)this installation guide: https://tutorials.ubuntu.com/tutorial/tutorial-create-a-usb-stick-on-windows#0

   3)create a bootable USB flash disk include ubuntu image, using Rufus :https://rufus.akeo.ie/

   on the empty linux comupter:

   1) boot the computer using the USB flash disk

   2) install linux

2.2 download project from github

https://github.com/udacity/CarND-Extended-Kalman-Filter-Project

2.3 install uWebSocket

uWebSocket is software to connect our C++ program and the simulator.

In the project folder, execute sudo sh ./install-ubuntu.sh

This command will installed uWebSockets.

2.4 download simulator

term2_sim.x86_64, download it from https://github.com/udacity/self-driving-car-sim/releases/

hins:

  1. add execute permission to the file sudo chmod -c 777 <your file name>
  2. The simulator may corrupt sometimes, so unzip it from the download file, or download it again

2.5 install anaconda and jupyter notebook

The python jupyter notebook is used for analyze the output log file.

check conda version

conda --version

If you computer don't have anaconda, then you could download anaconda or miniconda. Any of them is OK.

miniconda is about 50M. Download from:

https://conda.io/miniconda.html

anaconda is about 500M. Download from:

https://www.anaconda.com/download/#linux

Then install it,

  • miniconda

    bash Miniconda3-latest-Linux-x86_64.sh
  • anaconda

    bash Anaconda-latest-Linux-x86_64.sh

I use the enviroment from udacity Carnd-term1

https://github.com/udacity/CarND-Term1-Starter-Kit/blob/master/doc/configure_via_anaconda.md

Set the enviroment like this:

git clone https://github.com/udacity/CarND-Term1-Starter-Kit.git
cd CarND-Term1-Starter-Kit

Then run these commands, will create an conda enviroment, include the libraries we need.

conda env create -f environment.yml

If conda download these package too slow, you can use other channels instead. Edit your environment.yml file.

channels:

- https://conda.anaconda.org/menpo

- https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/

- https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/menpo/

- https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/

- conda-forge

Activate the enviroment

source activate carnd-term1

Then get into the jupyter notebook

jupyter notebook

3. Install some usefull software (optional)

3.1 Install chrome in ubuntu

   (If you like, the firefox is good enough also)

https://www.cyberciti.biz/faq/how-to-install-google-chrome-in-ubuntu-linux-12-xx-13-xx/

3.2 Install uGet download tools

If you want download something big in linux, you may need a download tool.

The uGet website: http://ugetdm.com/downloads-ubuntu

Install uget using command line

sudo add-apt-repository ppa:plushuang-tw/uget-stable
sudo apt update
sudo apt install uget

Then, install aria2 plugin for uGet

sudo apt install aria2

After these, search uGet in ubuntu search bar, you will found uGet.

3.3 Haroopad markdown editor

download from this site: http://pad.haroopress.com/user.html

Your downloaded Haroopad package could be an deb package. How to install an deb package?

sudo dpkg -i xxxx.deb

4.Data Flow

  1. The measuremennt processor/matlab simulator is generating the FUSION .txt file:

    "data/obj_pose-laser-radar-synthetic-ukf-input.txt";

    OR

    "../matlab_examples/obj_pose-laser-radar-synthetic-ukf-input.txt";

The Input file format is:

L(for laser) meas_px meas_py timestamp gt_px gt_py gt_vx gt_vy (gt is ground truth)

R(for radar) meas_rho meas_phi meas_rho_dot timestamp gt_px gt_py gt_vx gt_vy (gt is ground truth)

Example:

R 8.60363 0.0290616 -2.99903 1477010443399637 8.6 0.25 -3.00029 0

L 8.45 0.25 1477010443349642 8.45 0.25 -3.00027 0

  1. The EKF Algorithm reads form file reads all the lines and generates measurement structures

  2. The MeasurementProcessor() is called with individual measurements (one by one). The results are saved

    (Attention: no file processing routines are used inside MeasurementProcessor() all the file processing routines are in the main function.

    So the data read/write is decoupled from the algorithm

  3. The results are saved in an output file:

    "data/obj_pose-laser-radar-ekf-output.txt"

Output file format:

est_px est_py est_vx est_vy meas_px meas_py gt_px gt_py gt_vx gt_vy

Example:

4.53271 0.279 -0.842172 53.1339 4.29136 0.215312 2.28434 0.226323

43.2222 2.65959 0.931181 23.2469 4.29136 0.215312 2.28434 0.226323

5.CMakeLists.txt

There is nothing need to change in the CMakeLists.txtx

6.Program structure

overview of the formular

Tools.h

define a class named "tools" , and 2 method:

  • CalculateRMSE
  • CalculateJacobian

Tools.app

Implementation of the 2 method.

measurement_package.h

Defines a class called "MeasurementPackage".

It include the data of a measurement.

Kalman_filter.h

Defines a class named "KalmanFilter"

The class include some data

  • x_ state vector
  • P_ covariance matrix
  • F_ state transition matrix
  • Q_ process covariance matrix
  • H_ measurement matrix
  • R_ measurement covariance

The class defined some method

  • Init: initializes Kalman filter
  • Predict :predicts the state and state covariance, using the process model
  • update :update the state by using standard Kalman Filter equations.
  • updateEKF :update the state by using Extended Kalman Filter equations.

Kalman_filter.cpp

Implementation the method defined by Kalman_filter.h

FusionEKF.h

Head file of FusionEKF.cpp

Define a class called FusionEKF

Define a method ProcessMeasurement, a public KalmanFilter

FusionEKF.cpp

Implementation the method difined by FusionEKF.h

Main.cpp

  • Check_arguments: I add an option to the program, select fusionoronlyRadaroronlyLidar
  • hasData:
  • main:

       - write logfile, depend on the option you select.The logfile could be log_fusion.txt or log_onlyRadar.txt or log_onlyRidar.txt.

       - using uWS to communicate with Simulator.

       - calculate RMSE(root mean squre error) between estimations and ground truth.

       - Json is the message type, send to, or receive from the simulator.

       - extrac variables form istringstream, and put them into

       - Call ProcessMeasurment(meas_package) for Kalman filter

       -

7 Compile and run

  • mkdir build
  • cd build
  • cmake ..
  • make
  • ./ExtendedKF fusion
  • Download simulator term2_sim.app (if in OSX) and open it. Click play! bottom, select Project 1/2: EKF and UKF, and press Start bottom to start.

8. If your program can not run

Download the simulator again. Maybe the simulator is crash.

9. Analyze the Output data.

The analyze can be see in the build/analyz_log_file.ipynb. You can use jupyter notebook to open it.

This picture is the compare of RMSE. Obviously, the error is radar > lidar > fusion.

Outro

The course provide almost full of the code.

In my experience,

1)The hardest part is build a enviroment, at last I choose ubuntu linux platform.

2)If the simulator has no response, download it again, or extract it from the zip file.

Udacity carnd2 Sensor Fusion, Extended Karman Filter (English)的更多相关文章

  1. Sensor fusion(传感器融合)

    From Wikipedia, the free encyclopedia 来自维基百科,免费的百科Sensor fusion is combining of sensory data or data ...

  2. Kalman Filter、Extended Kalman Filter以及Unscented Kalman Filter介绍

    模型定义 如上图所示,卡尔曼滤波(Kalman Filter)的基本模型和隐马尔可夫模型类似,不同的是隐马尔科夫模型考虑离散的状态空间,而卡尔曼滤波的状态空间以及观测空间都是连续的,并且都属于高斯分布 ...

  3. Kalman filter, Laser/Lidar measurement

    You can download this project from https://github.com/lionzheng10/LaserMeasurement The laser measure ...

  4. (转) How a Kalman filter works, in pictures

    How a Kalman filter works, in pictures I have to tell you about the Kalman filter, because what it d ...

  5. 卡尔曼滤波—Simple Kalman Filter for 2D tracking with OpenCV

    之前有关卡尔曼滤波的例子都比较简单,只能用于简单的理解卡尔曼滤波的基本步骤.现在让我们来看看卡尔曼滤波在实际中到底能做些什么吧.这里有一个使用卡尔曼滤波在窗口内跟踪鼠标移动的例子,原作者主页:http ...

  6. probabilistic robotics_Kalman filter(一)

    码农生活告一段落,继续.... 多元正态分布 协方差矩阵,为正定对称矩阵.det表示行列式 协方差反应随机样本变量各分量之间的相关性. 当变量的假设模型不一致时,不适合用高斯滤波. 叠加高斯噪声的线性 ...

  7. LightOJ 1074 - Extended Traffic (SPFA)

    http://lightoj.com/volume_showproblem.php?problem=1074 1074 - Extended Traffic   PDF (English) Stati ...

  8. SQL Server Extended Events 进阶 3:使用Extended Events UI

    开始采用Extended Events 最大的阻碍之一是需要使用Xquery和XML知识用来分析数据.创建和运行会话可以用T-SQL完成,但是无论使用什么目标,数据都会被转换为XML.这个限制在SQL ...

  9. {ICIP2014}{收录论文列表}

    This article come from HEREARS-L1: Learning Tuesday 10:30–12:30; Oral Session; Room: Leonard de Vinc ...

随机推荐

  1. django导入 views.py

    网上找了各种教程..均无解....自己摸索出来了..分享 给大家... 首先INSTALLED_APP 加入 自己的 app pili,然后如下图 from App名称 import views 就 ...

  2. java从笨鸟到菜鸟

    链接别人的笔记 http://www.cnblogs.com/cugwx/p/3536840.html

  3. hive支持事务及单行操作 update delete

    测试环境  Hive 1.2.1000.2.6.0.3-8 set hive.support.concurrency=true; set hive.exec.dynamic.partition.mod ...

  4. java——单例模式

    这个教程写的真好:https://blog.csdn.net/mnb65482/article/details/80458571 1. 什么是单例模式? 单例模式确保某个类只有一个实例. 单例类必须自 ...

  5. battery-historian结果分析

    准备电量数据: 1. 断开adb服务 执行:adb kill-server 2. 开启adb服务 执行:adb start-server或者adb devices这两步操作的作用:adb作为一种连接的 ...

  6. Object类中有哪些方法

    equals().hashCode().wait().notify().notifyAll().toString().clone().getClass().finalize()

  7. Python操作列表

    1.List Python内置的一种数据类型是列表:list.list是一种有序的集合,可以随时添加和删除其中的元素. 比如,列出班里所有同学的名字,就可以用一个list表示: >>> ...

  8. 浅谈jrebel

    有个同事提高个jrebel的工具,提起tomcat的热部署方案. jrebel是一款收费的JVM级的热部署工具包. JVM级的热部署也就是说,可以不重启JVM,让修改或添加的类加载到JVM中. 加载器 ...

  9. 夜神模拟器连不上adb的解决办法

    解决办法: 1.任务管理器里看下,adb.exe以及nox_adb.exe这2个进程有没有在运行?有的话就结束掉 2.找到开发环境的SDK的目录和夜神模拟器的目录,将SDK\platform-tool ...

  10. 性能测试工具Jmeter07-Jmeter性能测试实战

    测试需求:测试20个用户访问www.baozhenart.com在负载达到30QPS时的平均响应时间 QPS:Query Per Second每秒查询率.是一台查询服务器每秒能够处理的查询次数.在因特 ...