MONITORING AN INTERSECTION USINGA NETWORK OF LASER SCANNERS

H. Zhao, J. Cui, H. Zha,Peking University

K. Katabira, X. Shao, R. Shibasaki,University of Tokyo

Background (1)

Analyzing and Monitoring the traffic behavior in an intersection

Efficiently and accurately COLLECTING the TRAFFICDATA in an INTERSECTION

Real-timely DETECTINGDANGEOUS SITUATIONS.

Background (2)• Vision-based methods suffer mainly on the

following difficulties– Occlusion– Computation Cost– Illumination Change

1. Restrict camera’s setting condition

e.g. the camera is required to be set on a tall position, monitoring intersection from the above

e.g. monitor vehicles of limited lanes, do not discriminate moving objects.

To solve the problems

2. Target on a simplified situation

ObjectiveThis research propose a novel system for monitoring and collecting detailed traffic data, with easy setting condition, in an environment of complicated traffic behavior, such as intersection, using a network of single-row laser scanner.

System ImageServer PC

#2

#1

Laser Scanner

Pedes.

#1

#2

Car

Car

Bicycle

Network

An image of integrated laser data

Laser Scanner

LD1LD3

LD2

LD2

LD3

LD1

Integrated frame Single laser frame

Single laser frameSingle laser frame

Safety zone

Safety zone

Road side

Road side

10m

Processing Flow

1. Data Integration

…

2. Detection

3. Tracking

4. Recognition

Server

Client

Network

Measure dataLaser Scanner1

BG Subtraction

Clustering

BG Generation

Measure dataLaser Scanner N

BG Subtraction

Clustering

BG Generation

Moving object

Processing ModulesServer

Data Integration

Detection

Tracking

Recognition

Moving object

Measure Data

BG SubtractionBG Generation

Clustering

Client Measure Data

BG SubtractionBG Generation

Clustering

…Laser Scanner 1 Laser Scanner N

Calibration, Synchronization

Network

Recognition

Tracking

Detection

Data Integration

Measure Data

BG SubtractionBG Generation

Clustering

Server

Client

Network

trajectory

Frame k

group

cluster

LD1 LD2 LD3

Data Structure

observation

state

Difficulties

Difficulty 1

The data of the same object might be spatially disconnected

Difficulties

The data of different objects might be spatially close

There are different kinds of objects, complicated motions

Task 1

We are not able to rely only on distance, size, shape, location etc for detection and traction.

We need to find some other clues to associate the data of the same object together.

Data Feature

u1

u2e c

sA car

Data Feature

v1

v2

v3

v4

uu

u

uu

uu

Data Feature

u

uu

u

u

u

Object Model

(v3, sv3)

len1

len2p (v2, sv2)(v4, sv4)

(v1, sv1)

(c, val) u

u

(v1=dirv)

dirv

The previous frame

The current frame

More sensors bring more confusion in data association

Difficulty 2

Task 2Data association is difficult when facing a

complicated environment such as an intersection, where different kinds of objects, different motion patterns exist.

Also, a large number of distributed sensors will bring more difficulties in data association.

A data association algorithm is required to tackle the confusions.

A range streamSc

an a

ngle

(with

in a

lase

r sca

n)

Time (A stream of laser scans)

a moving object

a static object

A simple clustering resultSc

an a

ngle

(with

in a

lase

r sca

n)

Time (A stream of laser scans)

Data Association Method

trajectory

k-1 k

group

cluster

LD1 LD2 LD3 LD1 LD2 LD3 LD3

New trajectory

1

2

3

Peking Univ.

Peking Univ.

Sports Center

Overhead Bridge

Laser Scanner

Experiment

Video Camera

40cm

激光扫描仪

SummaryWe propose a novel system for monitoring and collecting detailed traffic data, with easy setting condition, in an environment of complicated traffic behavior, such as intersection, using a network of single-row laser scanners.

Future Topics

• Accuracy evaluation• Recognition module• Fusing laser scanner with video camera