スライド 1

Download Report

Transcript スライド 1

A Method for Modeling of Pedestrian
Flow in the Obstacle Space
using Laser Range Scanners
Yoshitaka NAKAMURA†, Yusuke WADA‡,
Teruo HIGASHINO‡ & Osamu TAKAHASHI†
†Future
University Hakodate, JAPAN
‡Osaka University, JAPAN
Background
• Development of sensing technologies
– Various phenomena become able to be measured as
digital data
– Some services based on this measuring data can be
provided
• Pedestrian flow attracts attention
– Movement of the pedestrians
• For orientation of the services based on mobility pattern of
pedestrians
• For urban planning
• For control of pedestrian for refuge instructions
2011/9/19
IWIN2011
2
Purpose of Research
• To generate pedestrian flow model with high
accuracy and law cost
– Cost of measuring sensor
• Price of sensor is as low as possible
• Number of sensor is as little as possible
– Cost of data processing
• Information to use is as little as possible
2011/9/19
IWIN2011
3
Related Work
• Pedestrian flow detection
– Cameras
• By recognizing the images of pedestrians
• Disadvantage: Privacy, Setting cost, Angle of view
– RFID tag
• By tracking with RFID of each pedestrian
• Disadvantage: Cost of RFIDs
– Counting number of pedestrians
• By counting the passage number of pedestrians in each
gateway
• Disadvantage: Affected by occlusions
2011/9/19
IWIN2011
4
Laser Range Scanners(LRS)
• Measures the distance to an object from LRS
• Advantage
–
–
–
–
Fast scanning of wide area
Little probability to infringe pedestrian’s privacy
Small cost of calculation
(Simple tracking by the difference of data is possible)
• Disadvantage
– Easy to lose target objects by obstacles
– Difficult to measure all pedestrians completely
2011/9/19
IWIN2011
5
Measuring Data of LRS
• Measurement time
• Position coordinate of pedestrian
• ID of pedestrian assigned by simple tracking of
UTM-30LX
 Simple tracking of UTM-30LX
 Judge the same pedstrian from the difference of measured data and assign
ID to the pedestrian
 If tracking is succeed, pedestrian’s movement history can be found by
ID
 ID becomes extinct in the place where tracking of the pedestrian failed
 Even if the same pedestrian is found again, the other ID is assigned
2011/9/19
IWIN2011
6
Precedent Experiment(1/2)
• Conducted in “Whity Umeda”
– By synchronizing 4 LRSs(HOKUYO UTM-30LX)
– Measure the height of pedestrians’ waist[17]
[17] Kawata, Ohya, Yuta, Santosh and Mori, “Development of ultra small
lightweight optical range sensor system,” in Proc. of IROS2005
2011/9/19
IWIN2011
7
Precedent Experiment(2/2)
26m
33m
2011/9/19
IWIN2011
• Beige area is the
movable area of
pedestrians
• Orange columns
are LRSs
• Pink lines are
the measuring
laser of LRSs
8
Performance of Tracking
• Simple tracking of UTM-30LX could continue
only for a short time
– 30%~40% of IDs’ life times are only 1 second
• Pedestrians are hidden behind obstacles such as pillars
• Pedestrians are also hidden behind other pedestrians
• Some pedestrians are staying near LRS and become large
obstacles
=>The other approaches are needed for pedestrian
flow generation
2011/9/19
IWIN2011
9
Approach
• Pedestrian flow is often used in
– Trajectory analysis of customers in commercial facilities
– Pedestrian flow analysis for refuge instruction, etc.
• In such case, tendencies of the pedestrians’
movement are more important than actual
behaviors of pedestrians
– Accurate tracking is impossible
Pay attention to the change of
population density in the partial domain
of the measurement area
2011/9/19
IWIN2011
10
Proposed Method
• Generate the pedestrian flow model from the
population density
– Divide measurement area into some square
domains(cells)
– Calculate population density of each domain
– Generate flow model based on the change and
distribution of population density
2011/9/19
IWIN2011
11
Assumptions
1. Pedestrians move to the only adjacent cell (in
all directions) from the cell where
himself/herself is now
2. Pedestrians move from the entrance to the exit
without making a detour under Assumption 1
3. Pedestrian is measured only once in each cell
where he/she passes
2011/9/19
IWIN2011
12
Overview of Method
1. Dived the measurement area into some unit cells
2. Select the gateway cells
3. Suppose the route candidate between each 2
gateways
4. Calculate the population density of each cell
5. Calculate the number of sojourners
6. Estimate the route where the pedestrian passed
and its traffic
7. Determinate the direction of the flow
2011/9/19
IWIN2011
13
Supposition of the Route Candidate
• Suppose the route candidate
according to Assumptions 1
& 2 between each two
different gateway cells.
2011/9/19
IWIN2011
14
Calculation of the Population Density
Sojourners
3
• Count it up how many
pedestrians existed in each
cell for constant period of
time => Population Density
• Exclude pedestrians moving
at a speed less than minimum
speed as sojourners
– Minimum speed = 20cm/s
2011/9/19
IWIN2011
15
Estimation of the Route
 One route corresponds to one
pedestrian
• Decide a route candidate
passing a cell with high density
as a route
• Update the density data to the
data removed a decided route
• Repeat the process until the
population of all cells become
almost 0
2011/9/19
IWIN2011
16
Determination of Direction
• Detect the speed and direction
of pedestrians’ movement from
tracking data of UTM-30LXs
• Count the direction ratio of
pedestrians’ movement in each
cell on route candidates
• Determine the direction ratio of
route based on direction ratio
of pedestrian’s movement
2011/9/19
IWIN2011
17
Example of Generated Flow(1/2)
• Direction of arrow:
Direction of the
pedestrian flow
• Thickness of arrow:
Quantity of the
pedestrian flow
sojourners
4
sojourner
1
• Numbers in Circles:
Average number of
sojourners of same cell
for each 1 minute
sojourners
3
Moring
2011/9/19
IWIN2011
18
Example of Generated Flow(2/2)
• Direction of arrow:
Direction of the
pedestrian flow
• Thickness of arrow:
Quantity of the
pedestrian flow
sojournerssojourners
2
2
sojourners
3
• Numbers in Circles:
Average number of
sojourners of same cell
for each 1 minute
sojourners
6
Evening
2011/9/19
IWIN2011
19
Performance Evaluation
• Evaluate the generated model using scenario
data formed by MobiREAL based on the
number of pedestrians in the gateway cells
– Compare the pedestrian flow model generated by
proposed method with the pedestrian’s movement of
scenario data
2011/9/19
IWIN2011
20
Scenario Data
• Set the origin
and destination
point of each
node
• Generate a
realistic
movement
model using
MobiREAL
2011/9/19
IWIN2011
21
Agreement Rate
• Compare the flow model generated from the scenario
data by proposed method with the scenario data
• Agreement rate: Ratio of
the cell which the route
of generation model
passed is the same as
scenario data
Scenario data
Flow model
• Average agreement rate
of route in the whole
scenario is about 80.79%
Agreement rate: 67%
2011/9/19
IWIN2011
22
Discussion
• Accuracy of the measuring population density
– This method takes the average value of several minutes
• the tendency of the pedestrian flow does not change in a short time
• the disappearance of the pedestrian by the obstacle is keeping for an
instant
– Actually, the disappearing pedestrians exist for a long time
depending on the placement of the obstacles(pedestrians)
• It is necessary to consider about method to estimate
density by disappearance probability or different method
to calculate the density by flow quantity
2011/9/19
IWIN2011
23
Conclusion
• Proposed a method for modeling pedestrian flow in
the space such as underground shopping center
– Using Laser Range Scanners
– Using the change of population density
• Future work
– Different calculation method of density corresponding
to the real environment
– Comparison with the data which completely measured
the movement trace of each pedestrian
2011/9/19
IWIN2011
24