System on Chip (SoC) for Object/Pattern Recognition IP Core ...

brasscoffeeΤεχνίτη Νοημοσύνη και Ρομποτική

17 Νοε 2013 (πριν από 3 χρόνια και 9 μήνες)

69 εμφανίσεις

System on Chip (SoC) for
Object/Pattern Recognition IP Core


Project Presentation


Sampletalk Technologies Group


Ilya Kochubeevsky

Irena Kemarsky

Andrew Gleibman


Copyright © Sampletalk Technologies 2006
-
2012


INTRODUCTION


Existing security/alert systems
with computer vision elements,
e.g.,








typically are specialized and have narrow application range:




prevention of car theft (1) or fire (2);


detecting fatigue and alerting
drivers tending to sleep (3, 4);


recording road situations in driving time
(5);


recognizing vehicle departure from lane and danger of collision (6)
etc.


1)
Smart Car Security System

for detecting car thefts (with face recognition)

by Mr. K. Kalimuthu

2)
Fire Detection for Cargo compartment

by m
-
u
-
t Aviation
-
Technology

3)
Detecting Attention, Detecting Sleep

by Matthew Parks

4)
Smart Eye Pro

by SmartEye

5)
Mobile DVR

by FSEC

6)
Mobile Collision Prevention System

by MobileEye

Every such device has related cost and requires special
space in a vehicle or room where the device should function.


Combining several such devices in a single vehicle is
problematic for many reasons: space limit, signal
incompatibility etc.


On the other hand, the user might want a combined and more
versatile system, capable of recognizing diverse objects and
situations in real time. Furthermore, for any given system,
new situations may arise, which should be treated as singular
for rising an alert or for special purposes.


Building a separate security/alerting system for any such
situation and purpose is impractical.

PROPOSED METHOD


We introduce
a single low
-
cost multipurpose video
-
based
system that can be tuned for performing and combining most
of such tasks and more in real time. Furthermore, signals of
different nature and sources can be recognized by a single
system and joined into a simple alerting language.


The system is trainable for detecting various objects and
situations


on the road, in a production line, in working
environment etc.







Below examples of objects and situations, recognized and
detected by the current prototype of the system, are provided.

Detecting pre
-
contact

situation in using

electrical appliances

Detecting contact

situation

Below examples of objects and situations, recognized and
detected by the current prototype of the system, are provided.

Detecting intrusion

of human fingers

Below examples of objects and situations, recognized and
detected by the current prototype of the system, are provided.

Detecting assemblies

among simple parts

Below examples of objects and situations, recognized and
detected by the current prototype of the system, are provided.

Detecting
iris

and evaluating

degree of

openness of

human eye

Below examples of objects and situations, recognized and
detected by the current prototype of the system, are provided.

Detecting iris

and evaluating

degree of

openness

of

human eye

Below examples of objects and situations, recognized and
detected by the current prototype of the system, are provided.

Detecting vehicle

tires among

other objects

Below examples of objects and situations, recognized and
detected by the current prototype of the system, are provided.

Locating

vehicle number

plates for

subsequent

identification

Below examples of objects and situations, recognized and
detected by the current prototype of the system, are provided.

Detecting

intrusion

of human

hand


Below examples of objects and situations, recognized and
detected by the current prototype of the system, are provided.

All this detection has been done
using the
pre
-
prototype

of a SINGLE system




System on Chip (SoC) for
Object/Pattern Recognition IP Core





In the following scheme our SoC for Object/Patten Recognition is
shown as a part of some target system:



SoC

For object/
pattern

recognition


DDR2

Memory

Flash

PHY

System on Chip (SoC) for Object/Parren Recognition

System on Chip
(SoC) IP Core contains originally designed object/pattern recognition
algorithm Core. This algorithm requires training on typical objects or
situation patterns which should be recognized. SoC also contains
Soft Processor Core, Memory Interface Core and several other Cores
(Ethernet, Parallel etc). The chip has RGB parallel inputs for two or
more Image Sensors, several output channels for Signals and
Alarms, and a Touch Screen interface.


Embedded System

contains Proprietary chip (above), Memory chip
(DDR2 for example), Flash Memory chip, Ethernet Physical chip
(Phy) etc. Embedded system is a card with sockets for



-

Two or more cameras,


-

Signals, Alarms,


-

Touch Screen.






Target System
can be designed for Safety, Security and
Reliability in many areas. For instance, In Automotive area, the
following applications can be built:

-

car simulator,

-

combined system consisting of accident prevention, driver
fatigue prevention, smart "green box", tire defect recognition,
alerting about intrusions.