Harman-SEU Joint Lab Research Topics

soilflippantAI and Robotics

Nov 17, 2013 (3 years and 6 months ago)

109 views





1

|
P a g e


1 1/1 7/2 0 1 3


关于发布“东大

哈曼车联网技术联合实验室
2013
年度研发项目信息”的
通知

东大

哈曼车联网技术联合实验室是由东南大学与哈曼国际工业股份有限公司

2011
年共建的以车载信息技术为主的研发平台。根据中心发展需要,经双方论证
商定
2013
年研发计划,现面向全校范围征寻项目承担者,欢迎全校教师积极参与中
心项目研究。请感兴趣的老师

3

31
日之前
与中心
秘书
进一步联系,讨论项目具
体信息。谢谢!

研发项目信息参见附件。


联系人:唐利梅


东大

哈曼车联网技术联合实验室”秘书



机:
18662187646

E
-
mai
l:

jwribbon@163.com










Harman
-
SEU Joint Lab Research Topics


Tony Ma

Harman

Mar 6, 3013





2

|
P a g e


1 1/1 7/2 0 1 3







3

|
P a g e


1 1/1 7/2 0 1 3



Table of Contents

1.

High performance PHY of V2X

................................
................................
................................
.

4

2.

Precise position of WAVE

................................
................................
................................
..........

5

3.

Autonomous Driving based on data fusion of on
-
board sensors

................................
.........

6

4.

Online user comment
s filter

................................
................................
................................
.......

7

5.

Understanding driver visual behavior and fatigue status

................................
......................

8

6.

Driver’s emotion detection for song recommendation

................................
...........................

9

7.

Anti passenger coupling filtering for the automotive acoustic environment

.....................

10

8.

Machine learning on Listener’s preference on low l
evel sound setting

............................

11

9.

Internal Antenna Design for T
-
box

................................
................................
.........................

12

10.

Feasibility of stereoscopy of two fisheye cameras in dependence o
n camera solution

13

11.

A 3D
-
graphical simulation environment for stereoscopic or panoramic research

.......

14

12.

Integration of Irregu
lar Sparse 3D information for improvement of pseudo
-
3D
graphical model with video texturing

................................
................................
............................

15

13.

Magnesium material application for Car infotainment system
................................
........

16





REVISION HISTORY


Revision

#

Author

Revision Date

Reason

1.0

Jianjun Ma

Mar 6, 2013

initialization


















4

|
P a g e


1 1/1 7/2 0 1 3



1.

High performance PHY of V2X

Research description

Research on the high performance WAVE system for DSRC (dedicated short range
communi
cation), especially focuses on innovative technologies of the PHY and MAC
layers, to provide a more efficient and robust vehicular communication system, including
simulation of these technologies (software simulation, & road test) and its prototype.


Matri
x

Provide a high performance WAVE communication system which is efficient in PHY and
MAC layer:

1. Disruptive PHY technology to drive down BER in different SNR environment (lower
BER by more than 60% in typical 5.9GHz wireless communication environment);

2
. Compatibility: compatible with the 802.11p standard (communication with standard
802.11p devices);

3. Low latency: meet the requirement of DSRC;

4. Chipset integration: implement the system by hardware and integrate it into a
commercially available chips
et.



Plan

Apr. 2012~Mar.2013 lab research & simulation of the technology

Apr.2013~Sep.2013 prototype development

Oct. 2013~Mar.2014 field trial of the prototype

.

Evaluation Indicators

Research / solution reports

Patents (at least 8 accepted by Harman)

Si
mulation (source code, soft platform simulation report, field trial report)

Prototype (hardware / software platform and source code)





5

|
P a g e


1 1/1 7/2 0 1 3


2.

Precise position of WAVE

Research description

Research on IEEE 802.11p multi
-
carrier modulation based precision positioning

and
testing technology, which satisfies quick identification and location in emergency and
geographic information for routing in vehicular Ad Hoc networks).


Matrix

Provide the precise positioning solution and prototype based on WAVE systems, which is
fea
tured by its low latency, high accuracy and reliability.

1. Low latency: positioning time < 100ms;

2. High accuracy: positioning error <10cm;

3. Reliability: work in different environments, SNR=20dB.

4. Prototype: implement and integrate the solution into

a WAVE system.


Plan

Apr. 2013~Mar.2014 lab research & simulation of the technology

Apr.2014~Sep.2014 prototype development

Oct. 2014~Mar.2015 field trial of the prototype

.

Evaluation Indicators

Research / solution reports

Patents (at least 5 accepted by

Harman)

Simulation (source code and results)

Prototype (hardware / software platform and source code)







6

|
P a g e


1 1/1 7/2 0 1 3


3.

Autonomous Driving based on data fusion of on
-
board
sensors

Research description

Research on autonomous driving technology which enables autonomous dri
ving vehicles
and ensure their safety. All kinds of on
-
board sensors, such as cameras, radars,
communication modules (V2X, cellular, NFS, Bluetooth, WiFi …) will be used to collect
environment data and help to make driving decisions.


Matrix

Defining the d
estination, the system can drive the car to the destination autonomously,
safely and efficiently.


Plan

Apr. 2013~Mar.2014 lab research & simulation of the technology

Apr.2014~Sep.2014 prototype development

Oct. 2014~Mar.2015 field trial of the prototype

.

Evaluation Indicators

Research / solution reports

Patents (at least 10)

Simulation (source code and results)

Prototype (hardware / software platform and source code)










7

|
P a g e


1 1/1 7/2 0 1 3


4.

Online u
ser comments filter

Research description

A lot of internet applications hav
e user comment function, which provide user feedback to
the products, and also impact to other users.

We want to create the model to analysis the comments. It can filter the unmeaning
comments and provide the usefully comment to other users.

1) We can get

the comment from www.dianping.com or www.yelp .com by platform API.

2) Filter the unmeaning comments in the sample data.

3) Process and pick
-
up the usefully information to other user.


Matrix

1)

This can be embedded in different system.

2)

Low maintain
cost and Fast processing speed.

3)

The filter can identify the unmeaning information & usefully information as
human understanding.

a.

Keep all the comments are on the topic/products.

b.

Timeliness: The comments should be a useful experience.

c.

Filter has

the ability of self
-
learning to improve/adapt to difference application.

d.

The information is integrity as comments.

e.

No redundancy/repeated information.

f.

No Irrigation comments and the speed of Policy don’t allow.

4)

The model should be implemented.

5)

The accuracy should be greater than %95 comparing with human understanding.


Plan

Provides the design model and verifiability application in 6 months.

Evaluation Indicators

Provide integrated solution model and documents on time.

Provides emulation and

do the random and correct recognition rate should be greater than
95%.





8

|
P a g e


1 1/1 7/2 0 1 3


5.

Understanding driver visual
behavior

and fatigue status

Research description

Detail descript research content includes background, target and conditions.

In Modern times, drowsy drivin
g is one of the most severe factors caused traffic accident.
The accident not only brings serious problem and great effect to society, but also causes
hell and gone loss or property and precious lives. It is necessary and meaningful to study
the detection
of driver’s fatigue status and auxiliary drive device. Based on the face
detection, we study the application of the driver’s fatigue status. Through analyzing the
face video, system makes the right disposal according to driver

s fatigue status judged by
mo
uth and eyes. In order to improve the accuracy of the system, we introduce Lane
departure warning system, when driver is in fatigue status and Lane departure happens,
system will send out a warning, make a speed
-
down and flashlight, also active the
auxilia
ry drive device, drive the car to secure parking area.

Matrix

Detail and specify performance matrix data of this project

Rough
Plan

Step1: Design a face recognition algorithm

Step2: Design Lane departure warning system

Step3: Design module to automaticall
y get nearest secure parking area

Step4: Coding for face recognition algorithm

Step5: Coding for Lane departure warning system


Evaluation Indicators

Detail requirement of patent, report, demo, source code, material, design document, etc.

1.

High level d
esign document

2.

Face recognition algorithm module design document (here may produce a patent)

3.

Lane departure warning system module design document (here may produce a
patent)

4.

Code for above modules

5.

IT report, ST report

6.

Demo to integration tes
ters





9

|
P a g e


1 1/1 7/2 0 1 3


6.

Driver’s emotion detection for song recommendation

Research description

It is a usual experience that listeners have different preferences on songs in different mood,
f.x. sober, excited, moody, etc. This mood indicator provides a very important facto
r for
the interaction between the driver and automotive infotainment, f.x, allow the system to
recommend songs from a huge media library to the driver by learning the driver’s
preference on songs in different moods, or even recommend online contents.

By m
odern technologies, the driver’s mood could be detected by methods, such as face
recognition, driving behavior, blinking behavior analysis or more.

.

Matrix

Accuracy of the algorithm to detect driver’s mood. This Accuracy could be measured by a
few machin
e learning evaluation index and methods.

Complexity: More sensors? Hardware? Computation and load for the system, etc.


Plan

1.

Get extensive algorithmic research on mood detection and prepare the best
method candidates for the driving scenarios.

2.

Imple
ment the algorithm and integrate it to testing system which could be used in
the real automotive environment.

3.

Test this algorithm and provide it to be feasible for a commercial product


Evaluation Indicators:

1.

Complexity and feasibility are allowed

for the infotainment system.


2.

Patent and research papers have to be acknowledged by Harman


3.

Validation by trusty test data. It has to reach a commercial product level.






10

|
P a g e


1 1/1 7/2 0 1 3


7.

Anti passenger coupling filtering for the automotive
acoustic environment

Resea
rch description

An acoustic coupling phenomenon will take place when passengers sitting close in a
confined acoustic cabin such as the automotive cockpit. It triggers a minor frequency and
phase response change in the audio path. Different occupancy has di
fferent change.


Matrix

Measurement of the improvement given by this filtering will follow the automotive
acoustic measurement standard. The relevant party has to provide a stable improvement for
different car type, loudspeaker and occupancy scenarios.

C
omplexity: Computation cost for the system, etc plus acceptable improvement on
frequency and phase response.

.

Plan

1.

Get extensive algorithmic research on mood detection and prepare the best
method candidates for the driving scenarios.

2.

Implement th
e algorithm and integrate it to testing system which could be used in
the real automotive environment.

3.

Test
this algorithm

and provide it to be feasible for a commercial product


Evaluation Indicators

1.

Computation and complexity of the algorithm.D
SP algorithmic component
evaluation index.

2.

Acoustic experience improvement, objectively and subjectively.

3.

Robustness for different car type and occupancy.





11

|
P a g e


1 1/1 7/2 0 1 3


8.

Machine learning on Listener’s preference on low level
sound setting

Research description:

It

is always not a single sound effect that listeners are going after when they commit the
operations on the knob or panel of an audio system. For example , a listener choose to
boost the bass band in the EQ setting to hear more clearly on a bass instrument,

at the same
time ,he also might turn up the volume with the intention to raise the sound of those bass
instrument.

Listeners have the preferences on different sound effect f.x. loudness, dynamics (loudness
variance), brightness and roughness, timbre. The

preference on different low level sound
feature could be highly correlated, although how and why these factors correlated and how
it is decided by the listener still need to be figured out.

For example, if the user has the preference on hearing clearly o
n percussion sound, a
slower attack time and at the same time an EQ preference on the bass band would be
preferred at the same time, so to speak, correlated for the same preference.


Matrix

Accuracy of the algorithm to learn driver’s preference. This Acc
uracy could be measured
by a few machine learning evaluation index and methods.

Complexity: Computation and load for the system, etc.

Plan
:

1.

Get extensive algorithmic research on mood detection and prepare the best
method candidates for the driving sc
enarios.

2.

Implement the algorithm and integrate it to testing system which could be used in
the real automotive environment.

3.

Test
this algorithm

and provide it to be feasible for a commercial product

Evaluation Indicators:

1.

Complexity and feasibi
lity are allowed for the infotainment system.

2.

Patent and research papers have to be acknowledged by Harman

3.

Validation by trusty test data. It has to reach a commercial product level.





12

|
P a g e


1 1/1 7/2 0 1 3


9.

Internal Antenna Design for T
-
box

Research description

For Telemat
ics unit design, currently the antenna used in the car is externally mounted on
the roof. As the
Telematics

box shall have the E
-
call function which means to Call
emergency “110” or “119” if the car has the accident and broken down, but if the antenna
on t
he roof is damaged during the accident, the emergency call will not be successful.

So to improve the safety, it is better to design internal antenna as backup solution when
roof antenna is broken.

The internal antenna is for phone call, should support 2G
/3G and 4G in future.

Space for the antenna is 60mm*14*18.




Matrix


Frequency Bands: 2G, 3G and 4G all bands

Impedance: 50 ohm

VSWR min. 3:1

Efficiency 700
-
800MHz >35%;

ALL others bands >50%.


Plan

Concept
---

May, 2013

Spec


Jun, 2013

Design and simulation


July, 2013

Mockup and

verification


Sep, 2013





13

|
P a g e


1 1/1 7/2 0 1 3


.

Evaluation Indicators

1)

Concept description

2)

Simulation Result

3)

Design files

4)

Test report.

10.

Feasibility of stereoscopy of two fisheye cameras in
dependence on camera solution

T
he Project Goal:

Study of stereoscope of two fisheye cameras mounted with large offset and non
-
parallel.
The typical camera mounting is in the surround view system where for example the front
and the side cameras are nearly perpendicular to each other with

an offset of more than one
meter. Such camera mounting results in a small overlapping of the adjacent camera views.
With such a camera configuration, the work is to verify if the stereoscope is still feasible
with VGA cameras or megapixel cameras. If not
feasible, in what condition is such camera
mounting still realizable, say increasing camera resolution etc.

Inputs
:

1.

Simulation environment and/or camera samples (CVBS cameras or megapixel
cameras). The Simulation Environment should be available. Camer
a samples are optional.

2.

In case of simulation environment, specify a camera mounting similar to a
surround view configuration. If real camera samples are used, Harman and SEU need to
work together to set up a camera rig and capture videos/images for fur
ther investigation.

Outputs
:

1.

Knowledge if the stereoscopy is achievable in the specified camera configuration
and the proof for that including uncertainty analysis etc.

2.

Depth information if achievable with uncertainty indication.

3.

Suggestion of a
feasible stereoscopy of fisheye cameras

Projection
:

Suggest the effort and the schedule.





14

|
P a g e


1 1/1 7/2 0 1 3


11.

A 3D
-
graphical simulation environment for stereoscopic
or panoramic research

The Project Goal:

Set up a 3D graphical environment simulating vehicle environment like pa
rking scene or
highway scene for creating simulative videos captured by four or more cameras with
configurable intrinsic and extrinsic parameters.


Inputs
:

1.

Specification of the environment in physical dimension regarding the position &
orientation of t
he ego
-
vehicle and the surrounding objects (other vehicles, obstacles like
poles, walls etc.), parking marks, lane marks, etc. in a user comprehensible format like xml,
json or else.

2.

Specification of the camera intrinsic (according to a fish eye model)
and extrinsic
parameter (camera mounting parameters) in a user comprehensible format like xml or else.

3.

Specification of dynamic motion models for the ego car and other moving objects
like other cars

4.

Specification of the video format and length


Outpu
ts
:

1.

A 3D graphical viewer with all specified objects. When the simulation starts, the
viewer shows the dynamics of the whole scene.

2.

Simulative videos captured by each video camera with synchronized time stamps.


Projection
:

Propose a modeling tool
such as Matlab VR toolbox, Blender or something else.

Suggest the effort and the schedule.





15

|
P a g e


1 1/1 7/2 0 1 3


12.

Integration of Irregular
Sparse

3D information for
improvement of pseudo
-
3D graphical model with video
texturing

The Project Goal:

Due to lack of 3D information, th
e regular surround view system with four or more
cameras assumes a pseudo
-
3D environment such as an ellipsoid mimicking the real
-
world
environment of the ego car. Then the video are statically stitched as steaming textures to
this graphical object to creat
e an immersive virtual reality. This only works for the part of
the stitching where the assumption meets the real depth such as textures on the floor. For
most part of the stitching, since the graphical model is static and the real environment is
dynamic,
the deviation of the assumed depth from the real value is almost always there, so
that the stitching mostly only works from a top view.


Utilization of other sensors like Radar or stereoscopy can provide sparse 3D information
for dynamically modifying the
graphical model so that it approximates the real world
better that a static graphic model. In consequence, the stitching quality will be improved.


Thus the work is to find an optimal way of integrate sparse 3D information which can be
provided by other se
nsors like radar or stereoscopy into video texturing process to improve
the graphical model and thus the stitching quality.


Inputs
:

1.

Synchronized video streams captured either by simulation or by real camera
samples.

2.

Sparse 3D points of the environm
ent acquired either by simulation or manual
annotation.

3.

Specification of an initial graphical object like an ellipsoid with given
triangulation and vertices.


Outputs
:





16

|
P a g e


1 1/1 7/2 0 1 3


1.

Dynamically modified graphical model with updated triangulation and new
vertices.


2.

The result should be validated by simulation or real camera images before and
after the integration of sparse 3D information.


Projection
:

Suggest the effort and the schedule.

13.

Magnesium

material application for Car infotainment
system

Research descrip
tion

Background:

GM need module with light weight, good EMC performance, good physical
characteristics

Target: To

find out a kind of material to fulfill customer requirements and meet our design.

C
onditions:

salt spray:

96h


Storage t
emperature:
-
30
-
90 degree


High humidity:90% under temperature:65 degree.

.

Matrix

-
need a kind of material with lower density almost same as Mg alloy AZ91D, good
oxidation proof
characteristics (
or with second process treatment), good

physical
characteristics

such as
strength, stiffness

etc…

-
Material surface electrical resistance shall be less than 0.1Ω after second
process, high

temperature storage
test, high

humidity storage
test. Salt

spray.

-
0.1Ω measurement method:
-
measuring procedure with brass inspection
pieces(φ40mm×height 30mm)


-
measureing distance of axis 50mm on plan surface.





17

|
P a g e


1 1/1 7/2 0 1 3





Plan

2014/12
mass manufacture ready

.

Evaluation Indicators

Detail requirement of patent, report, demo, source code, material, design document, etc.