Sparse Mat: A Tale of Devising A Low-Cost Directional System for Pedestrian Counting

Sparse Mat: A Tale of Devising A Low-Cost Directional System for Pedestrian Counting

Department of Computer Science and Engineering,Bangladesh University of Engineering and Technology, Dhaka, Bangladesh

NSysS 2017Dhaka, Bangladesh

Tarik Reza Toha, Salman Estyak, Taslim Arefin Khan, Tusher Chakraborty, and A. B. M. Alim Al Islam

2

Outline

• Background and motivation• Our proposed system– Design– Implementation– Experimentation

• User evaluation• Conclusion and future work

3

Background: Pedestrian Counting

4

Applications of Pedestrian Counting

• Strategic business research – Tourist flow estimation

• Security and emergency support– Victim count estimation

for rescue operations• Optimization tasks– Railway stations,

shopping malls, airports, hospitals, etc.

Assign staff deployment to demand

5

Classical Pedestrian Counting Technique

• Low human mobility– Using pen and paper,

or tally counter

• High human mobility– Automated pedestrian counter is

required

?

6

Existing Automated Counting Techniques

Pros: High accuracy

Cons: High cost High power

Pros: Low-cost Low-power Easy setup process

Cons: Low accuracy

Laser scanner counter[Katabira et al., ACRS, 2004]

Infrared beam counter[Wienand et al., USPTO, 2001]

Cons: High cost Complex algorithms High processing power

Pros: High accuracy

Camera-based counter[Agusta et al., ICCEMS, 2014]

Existing high-end device based solutions are high-cost

and high-power

7

Existing Mat-Based Counting Technologies

Greneker et al., 1996(2000 sensors/sq. ft.)

Kutschera et al., 2011(90 sensors/sq. ft.)

Instant Counting Mat by Milon et al., 2013(56 sensors/sq. ft.)

Existing mat-based solutions demand high densities of

expensive sensors

8

Our Contribution

We propose a mat-based pedestrian counting solution comprising low density (~3 sensors/sq. ft.) of low-cost sensors, which exhibits a high accuracy in pedestrian

counting

Sparse Mat

9

Underlying Sensing Mechanism

10

Sparse Mat: Operational Block Diagram

Block diagram

Cross sectional view of sensing module

11

Snapshot of Sparse Mat

Sensing Module Controller Module

12

Piezoelectric Sensor Responses for Varying Radial Distances

Piezoelectric sensor responses under varying points of pressure over the hardboard block

Coverage of piezoelectric sensor response

Vibration propagation while knocking on a hardboard block, which covered a

piezoelectric sensor

13

Sensing Step-Down Phenomena

Sensor placement over mat Sensor placement along with taped foam

Hardboard block placement over sensors

14

Findings from 2×2 Piezo-grid

Piezoelectric sensor responses while moving towards from down row to up row over the left sensors

Piezoelectric sensor responses while moving towards from up row to down row over the right sensors

drdl

ul ur

drdl

ul ur

The varying response time can significantly undermine the accuracy of directional

pedestrian counting

15

Resilience to Variation in Response Time: 2×3 Piezo-grid?

Sensor placement over mat Sensor placement along with taped foam

Hardboard block placement over sensors

16

Findings from 2×3 Piezo-grid

Piezoelectric sensor responses while moving towards from down row to up row

Piezoelectric sensor responses while moving towards from up row to down row

drdl

ul ur

dm

um

drdl

ul ur

dm

um

The ending always maintains perfect order even though

starting may not!

17

Sensing Step-up Phenomena

18

Modified Settings in 2×3 Piezo-grid

Sensor placement over mat Sensor placement along with taped foam

Hardboard block placement over sensors

19

Enhancing Sensing Accuracy: 3×2 Piezo-grid

Sensor placement over mat Sensor placement along with taped foam

Hardboard block placement over sensors

20

User Evaluation of Preliminary Designs

Grid Sensing Phenomena Accuracy (%) Causes

2×2 Step-Down < 50 Response time varies significantly from sensor to sensor

2×3 Step-Down 60 Time delay between placing heel and toe is very small

2×3 Step-Up 75 Participants often failed to place their foot step over both hardboard blocks

21

User Evaluation of the Final Design: 3×2 Piezo-grid Sensing Step-Up

Age Height (inch)

Weight (kg)

Shoe length (inch) Total attempts Successful

detectionDevice

accuracy (%)22 66 68 11 18 14 7822 70 62 11.25 20 20 10022 73 81 12 10 8 8023 70 90 11.25 20 20 10025 68 72 11.25 20 20 10026 66 55 11 8 7 88

Total 96 89 93

22

Conclusion and Future Work

• Existing automated directional pedestrian counters are generally highly expensive

• We propose a low-cost and easily-to-deploy automated directional pedestrian counter without compromising accuracy– Uses sparser sensors than any other state-of-the-art technologies

94% sparser than Instant Counting Mat

– User evaluation of real implementations confirms an average accuracy of 93%

• Future work– Workable for different types of users such as kids– Determine multiple footsteps of a single person– Determine multiple persons traversing in parallel over the sensor mat

A Short Demonstration of Sparse Mat

23

• https://youtu.be/C7MB58Xvr_w

Thank youQuestions are welcome!

Email: [email protected]

24

25

Variable Sensor Response Times

26

Towards Sensing Step-up Phenomena (contd.)

Flowchart of step-down phenomena sensing algorithm

Flowchart of step-up phenomena sensing algorithm