Developing Video Signal
Processing Algorithms for
Embedded Vision
Systems
Shogo MURAMATSU
Dept. of Elec. & Electronic Eng.
Niigata University
Contents
 Introduction to Embedded Vision
 Tutorial Material with MATLAB®/Simulink ®
EmbVision Tutorial
 Example Activities
Smart Dormitory Program (PBL Example)
NSOLT Project (Research Example)
 Conclusions
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
 Introduction to Embedded Vision
 Tutorial Material with MATLAB®/Simulink®
EmbVision Tutorial
 Example Activities
Smart Dormitory Program (PBL Example)
NSOLT Project (Research Example)
 Conclusions
The IoT Era is Beginning.
 Diversification of Sensing Environment
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Embedded Systems will be Deployed Broadly.
Seeing is Believing.
 Rapid Proliferation of Vision Systems
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Embedded
System
Computer
Vision
Embedded
Vision
Embedded Vision Systems gather Attention.
Contents
 Introduction to Embedded Vision
 Tutorial Material with MATLAB ® /Simulink ®
 EmbVision Tutorial
 Example Activities
 Smart Dormitory Program (PBL Example)
 NSOLT Project (Research Example)
 Conclusions
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
From DSP First to SI First
 DSP First: A Multimedia Approach (1998)
 Presents basic DSP concepts in an intuitive
style by multimedia signals with MATLAB.
 Evolution of MATLAB makes the System
Integration (SI) First Approach available.
 Embedded System Development
 Web + Database Management
 Network Connection
DB
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
EmbVision Tutorial
 Educational Material
for new members in
our Lab.
(Junior or Senior)
 Estimated to take 12H
SISA 2015 @ Chiba Institute of Tech.2015/8/26
Today, condensed to
50min!
http://msiplab.eng.niigata-u.ac.jp/embvision/en/
Demo with Raspberry Pi™
 Gradient Filtering on
Raspberry Pi
(Final Exercise)
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Overview of EmbVision Tutorial
 One can implement user-defined System
objects on Raspberry Pi.
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Part 1 Part 2 Part 3 Part 4 Part 5 Part 6
Intensity
Adjustment
Color space
Conv.
Grad.
Filter
System
Object™
Video
Stream
Process.
MATLAB
System
block
Unit
Test
Raspberry
Pi
MATLAB Simulink
Part 1:
Image I/O and Pixel Processing
 Target
 Read, display, and write images
 Simple pixel processing
 Exercises
 Intensity Adjustment
 Color Space Conversion
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
RGB2
HSV
HSV2
RGB
𝑆 ← 2𝑆
Part 2:
Filtering and Frequency Analysis
 Target
 1-D signal and 2-D image filtering
 Frequency analysis
 Exercises
 Horizontal Differential Filter
 Magnitude and Direction of Gradient
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Part 3:
Class Definition and Unit Testing
 Target
 Object-oriented programming in MATLAB
 Unit testing framework in MATLAB.
 Exercises
 HSV2RGB Class
 Gradient Filter Class
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Part 4: Video Stream Processing
- MATLAB -
 Target
Read, display and write videos in MATLAB
Video stream processing in MATLAB
 Exercises
Sobel Gradient Filter
Frame Difference
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Part 5: Video Stream Processing
- Simulink -
 Target
 Read and display videos with Simulink
 Use MATLAB System blocks on Simulink
 Video stream processing with Simulink
 Exercises
 Prewitt Gradient Filter
 Sobel Gradient Filter
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Part 6: Video Stream Processing
- Raspberry Pi™ -
 Target
 Simulate Simulink model in external mode
 Deploy Simulink model on Raspberry Pi
 Exercises
 Sobel Gradient Filter
 Deploy a created new model
(Option)
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
DIY
 List of Expected Extension Examples
 Raspberry Pi Camera Module Control
 Fixed-point Implementation
 Parallel Implementation (for Quad Core on RasPi2)
 Detection & Recognition of Objects
 Speech & Audio Signal Processing
 GPIO/I2C Controls
 Network Application Development
 etc.
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Contents
 Introduction to Embedded Vision
 Tutorial Material with MATLAB®/Simulink®
EmbVision Tutorial
 Example Activities
Smart Dormitory Program (PBL Example)
NSOLT Project (Research Example)
 Conclusions
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Smart Dormitory Program
 Project-Based Learning (PBL) Program at Facul.
of Eng., Niigata Univ.
 Financially Supported by MEXT
 MEXT: Ministry of Education, Culture, Sports, Science and
Technology
 15 Research Groups are working on their research
activities.
 The 2nd group is Cyber Physical System (CPS)
Dormitory
 Members are from different grades and departments.
 Develop an IoT App. with embedded vision systems
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
CPS Dormitory
 Develops a System for Estimation and Visualization of
Customer Congestion at a Student Cafeteria
 BeagleBone Black boards as Sensor Nodes
 ThingSpeak as an IoT data collection plathome
Installation of a Sensor Node Deployment Map of Sensor Nodes
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Presentation at the 4th MEXT
Science Inter-College
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
@Kobe Convention Center
29th Feb. - 1st Mar., 2015
 Nonseparable Oversampled Lapped Transform
 Sparsity-Aware Image and Volume Data Restoration
(a)
(b)
NSOLT Project
[Muramatsu,ICASSP2014]
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Simulink Model
Zynq® Implementation
 Xilinx®’s All Programmable SoC
ARM® CPU and FPGA are connected to each
other through AXI4 interface on a single chip
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Xilinx’s Zynq Inside of Zynq
Workflow of Co-implementation
Generation of HDL
(HDLCoder™)
Generation of C
(Embedded Coder® )
Configuration Build
Zynq
2015/8/26
System Model of NSOLT
(MATLAB/Simulink)
Isolation of HW and SW in the model
SISA 2015 @ Chiba Institute of Tech.
Contents
 Introduction to Embedded Vision
 Tutorial Material with MATLAB®/Simulink®
EmbVision Tutorial
 Example Activities
Smart Dormitory Program (PBL Example)
NSOLT Project (Research Example)
 Conclusions
2015/8/26 SISA 2015 @ Chiba Institute of Tech.
Conclusions
 This tutorial explained
 How to develop and evaluate video signal
processing algorithms with MATLAB/Simulink
 How to implement visual applications on
embedded systems with MATLAB/Simulink
 Two example activities were introduced
 CPS Dormitory Prog. as an Educational Activity
 NSOLT Project as a Research Activity
2015/8/26 SISA 2015 @ Chiba Institute of Tech.

Developing Video Signal Processing Algorithms for Embedded Vision Systems

  • 1.
    Developing Video Signal ProcessingAlgorithms for Embedded Vision Systems Shogo MURAMATSU Dept. of Elec. & Electronic Eng. Niigata University
  • 2.
    Contents  Introduction toEmbedded Vision  Tutorial Material with MATLAB®/Simulink ® EmbVision Tutorial  Example Activities Smart Dormitory Program (PBL Example) NSOLT Project (Research Example)  Conclusions 2015/8/26 SISA 2015 @ Chiba Institute of Tech.  Introduction to Embedded Vision  Tutorial Material with MATLAB®/Simulink® EmbVision Tutorial  Example Activities Smart Dormitory Program (PBL Example) NSOLT Project (Research Example)  Conclusions
  • 3.
    The IoT Erais Beginning.  Diversification of Sensing Environment 2015/8/26 SISA 2015 @ Chiba Institute of Tech. Embedded Systems will be Deployed Broadly.
  • 4.
    Seeing is Believing. Rapid Proliferation of Vision Systems 2015/8/26 SISA 2015 @ Chiba Institute of Tech. Embedded System Computer Vision Embedded Vision Embedded Vision Systems gather Attention.
  • 5.
    Contents  Introduction toEmbedded Vision  Tutorial Material with MATLAB ® /Simulink ®  EmbVision Tutorial  Example Activities  Smart Dormitory Program (PBL Example)  NSOLT Project (Research Example)  Conclusions 2015/8/26 SISA 2015 @ Chiba Institute of Tech.
  • 6.
    From DSP Firstto SI First  DSP First: A Multimedia Approach (1998)  Presents basic DSP concepts in an intuitive style by multimedia signals with MATLAB.  Evolution of MATLAB makes the System Integration (SI) First Approach available.  Embedded System Development  Web + Database Management  Network Connection DB 2015/8/26 SISA 2015 @ Chiba Institute of Tech.
  • 7.
    EmbVision Tutorial  EducationalMaterial for new members in our Lab. (Junior or Senior)  Estimated to take 12H SISA 2015 @ Chiba Institute of Tech.2015/8/26 Today, condensed to 50min! http://msiplab.eng.niigata-u.ac.jp/embvision/en/
  • 8.
    Demo with RaspberryPi™  Gradient Filtering on Raspberry Pi (Final Exercise) 2015/8/26 SISA 2015 @ Chiba Institute of Tech.
  • 9.
    Overview of EmbVisionTutorial  One can implement user-defined System objects on Raspberry Pi. 2015/8/26 SISA 2015 @ Chiba Institute of Tech. Part 1 Part 2 Part 3 Part 4 Part 5 Part 6 Intensity Adjustment Color space Conv. Grad. Filter System Object™ Video Stream Process. MATLAB System block Unit Test Raspberry Pi MATLAB Simulink
  • 10.
    Part 1: Image I/Oand Pixel Processing  Target  Read, display, and write images  Simple pixel processing  Exercises  Intensity Adjustment  Color Space Conversion 2015/8/26 SISA 2015 @ Chiba Institute of Tech. RGB2 HSV HSV2 RGB 𝑆 ← 2𝑆
  • 11.
    Part 2: Filtering andFrequency Analysis  Target  1-D signal and 2-D image filtering  Frequency analysis  Exercises  Horizontal Differential Filter  Magnitude and Direction of Gradient 2015/8/26 SISA 2015 @ Chiba Institute of Tech.
  • 12.
    Part 3: Class Definitionand Unit Testing  Target  Object-oriented programming in MATLAB  Unit testing framework in MATLAB.  Exercises  HSV2RGB Class  Gradient Filter Class 2015/8/26 SISA 2015 @ Chiba Institute of Tech.
  • 13.
    Part 4: VideoStream Processing - MATLAB -  Target Read, display and write videos in MATLAB Video stream processing in MATLAB  Exercises Sobel Gradient Filter Frame Difference 2015/8/26 SISA 2015 @ Chiba Institute of Tech.
  • 14.
    Part 5: VideoStream Processing - Simulink -  Target  Read and display videos with Simulink  Use MATLAB System blocks on Simulink  Video stream processing with Simulink  Exercises  Prewitt Gradient Filter  Sobel Gradient Filter 2015/8/26 SISA 2015 @ Chiba Institute of Tech.
  • 15.
    Part 6: VideoStream Processing - Raspberry Pi™ -  Target  Simulate Simulink model in external mode  Deploy Simulink model on Raspberry Pi  Exercises  Sobel Gradient Filter  Deploy a created new model (Option) 2015/8/26 SISA 2015 @ Chiba Institute of Tech.
  • 16.
    DIY  List ofExpected Extension Examples  Raspberry Pi Camera Module Control  Fixed-point Implementation  Parallel Implementation (for Quad Core on RasPi2)  Detection & Recognition of Objects  Speech & Audio Signal Processing  GPIO/I2C Controls  Network Application Development  etc. 2015/8/26 SISA 2015 @ Chiba Institute of Tech.
  • 17.
    Contents  Introduction toEmbedded Vision  Tutorial Material with MATLAB®/Simulink® EmbVision Tutorial  Example Activities Smart Dormitory Program (PBL Example) NSOLT Project (Research Example)  Conclusions 2015/8/26 SISA 2015 @ Chiba Institute of Tech.
  • 18.
    Smart Dormitory Program Project-Based Learning (PBL) Program at Facul. of Eng., Niigata Univ.  Financially Supported by MEXT  MEXT: Ministry of Education, Culture, Sports, Science and Technology  15 Research Groups are working on their research activities.  The 2nd group is Cyber Physical System (CPS) Dormitory  Members are from different grades and departments.  Develop an IoT App. with embedded vision systems 2015/8/26 SISA 2015 @ Chiba Institute of Tech.
  • 19.
    CPS Dormitory  Developsa System for Estimation and Visualization of Customer Congestion at a Student Cafeteria  BeagleBone Black boards as Sensor Nodes  ThingSpeak as an IoT data collection plathome Installation of a Sensor Node Deployment Map of Sensor Nodes 2015/8/26 SISA 2015 @ Chiba Institute of Tech.
  • 20.
    Presentation at the4th MEXT Science Inter-College 2015/8/26 SISA 2015 @ Chiba Institute of Tech. @Kobe Convention Center 29th Feb. - 1st Mar., 2015
  • 21.
     Nonseparable OversampledLapped Transform  Sparsity-Aware Image and Volume Data Restoration (a) (b) NSOLT Project [Muramatsu,ICASSP2014] 2015/8/26 SISA 2015 @ Chiba Institute of Tech. Simulink Model
  • 22.
    Zynq® Implementation  Xilinx®’sAll Programmable SoC ARM® CPU and FPGA are connected to each other through AXI4 interface on a single chip 2015/8/26 SISA 2015 @ Chiba Institute of Tech. Xilinx’s Zynq Inside of Zynq
  • 23.
    Workflow of Co-implementation Generationof HDL (HDLCoder™) Generation of C (Embedded Coder® ) Configuration Build Zynq 2015/8/26 System Model of NSOLT (MATLAB/Simulink) Isolation of HW and SW in the model SISA 2015 @ Chiba Institute of Tech.
  • 24.
    Contents  Introduction toEmbedded Vision  Tutorial Material with MATLAB®/Simulink® EmbVision Tutorial  Example Activities Smart Dormitory Program (PBL Example) NSOLT Project (Research Example)  Conclusions 2015/8/26 SISA 2015 @ Chiba Institute of Tech.
  • 25.
    Conclusions  This tutorialexplained  How to develop and evaluate video signal processing algorithms with MATLAB/Simulink  How to implement visual applications on embedded systems with MATLAB/Simulink  Two example activities were introduced  CPS Dormitory Prog. as an Educational Activity  NSOLT Project as a Research Activity 2015/8/26 SISA 2015 @ Chiba Institute of Tech.