FAST PROTOTYPING OF EMBEDDED IMAGE PROCESSING APPLICATION ON - - PowerPoint PPT Presentation

FAST PROTOTYPING OF EMBEDDED IMAGE PROCESSING APPLICATION ON HOMOGENOUS SYSTEM

1

• Hanen Chenini, Loïc Sieler, Mohamed Amine Boussadi, Thierry Tixier, Jean-Pierre Dérutin,

Alexis Landrault and Romuald Aufrére

`

• 1. Context
• 2. Real time parallel implementation on SoPC

(System on Programmable Chip)

• 3. Model-driven approach for real-time road

recognition

• 4. Experiments && Conclusion

2

`

3

` CMOS imager Embedded treatment SoC approach

• 1. Context

4

`

• !
• !"!

#

• 1. Context

5

`

6

`

µBlaze Soft Core Mem hardware Routeur µBlaze Soft Core hardware Routeur Mem Number of links=dimension of HyperCube

• 2. Real time parallel implementation on SoPC (System on Programmable Chip)

7

Homogeneous Network of Communicating Processors on SoPC. Parametrization of the HNCP via the CubeGen Framework (dimension, communication ,Size of memory, Configurations of softcore…). The generic architecture model chosen is the type MIMDDM. This parametrizable architecture is regular and homogeneous. Direct point to point links and the other using a hardware router. The static topology choice is the hypercube.

`

$ #

#

• %&

'

ISE/EDK

()

%*

• ##
• *
• +

%' ,

,

• 2. Real time parallel implementation on SoPC (System on Programmable Chip)
• #

8

`

9

`

10

`

11

• The steps contained in our proposed recognition process are :

Learning phase Recognition phase.

• 3. Model-driven approach for real-time road recognition

`

• 3. Model-driven approach for real-time road recognition

12

The model used is based on image data and camera parameters. This method is based on recursive recognition driven by a probabilistic model of the road edges in the image. Our statistical model is composed of n image parameters. This model is represented by a vector and its covariance matrix :

`

13

• 3. Model-driven approach for real-time road recognition

`

14

Regions of Interest

RoI

Regions of Interest

• RoI

Score of recognition 3

3 4 5 1 2

Score of recognition 1

1 2 3 4 5

Hypothesis 1 Score of recognition 2

2 3 4 5 1

Hypothesis 2 Hypothesis 3

• 3. Model-driven approach for real-time road recognition

`

15

Learning phase Definition of the best interest zone Detection step Detection? Updating Road Recognition NO YES

• 3. Model-driven approach for real-time road recognition

`

16

`

17

`

18

`

19

`

20

Road recognition Demo

`

21

• Some research of Institut Pascal groups focus their work in development of

autonomous robot navigation.

• As a consequence, we propose :
• !"#$%&'(

`

22