1 of 21

A picture is worth a thousand measurements: �a low-cost camera-based sensor �for in-situ particle size estimation

Canwei (Steven) Pang1, Baiqian Shi1, Stephen Catsamas1, David McCarthy1,2

1 BoSL Water Monitoring and Control, Department of Civil Engineering, Monash University, Wellington Road, Clayton, Victoria 3800, Australia

2 School of Civil and Environmental Engineering, Queensland University of Technology, S Block, Level 7, S727, Brisbane, Queensland 4000, Australia

*Corresponding author email: david.mccarthy@qut.edu.au

16th International Conference on Urban Drainage 2024 - Session 5.2b

1

Monash University & Queensland University of Technology

2 of 21

Why do we measure particle size?

  • Sediment a serious issue in stormwater
    • Sediment of different sizes behave differently.
  • Turbidity is a common tool to measure sediment as a surrogate.
    • But it only provides the big picture instead of the sizes of sediment in our stormwater system.
  • Particle size distribution (PSD) analysis
    • Water sampling + storage and transport + lab analysis
    • Hugely uncertain, laborious, expensive, and delayed information
    • How about making a sensor for PSD overcoming these issues?

2

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

3 of 21

The Simple Principle

3

Light Source

Turbidity, Total Suspended Solids (TSS), Water Clarity, and now for PSD…

A particle or whatever opaque

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

4 of 21

Project Aim

4

To develop, test, and validate a low-cost, real-time, in-situ sensor for estimating particle size distribution (PSD) of sediment within stormwater system.

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

5 of 21

Methodology – Experimental Setup

5

Clear acrylic plate

White LED light module

3d-printed, openable testing container (black)

Camera

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

6 of 21

Methodology – Experimental Setup

6

3d-printed, openable testing container (black)

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Testing sample with artificial particles

Camera

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

7 of 21

Methodology – Experimental Setup

7

3d-printed, openable testing container (black)

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Camera

Window for image capturing

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

8 of 21

Methodology – Image Processing Method

8

Step 1: Crop and convert original RGB image to grayscale image (0-255, from black to white).

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

9 of 21

Methodology – Image Processing Method

9

Step 2: Find a primary (and currently subjective) grayscale threshold which distinguish dark particles and bright background. Then apply it to create a binary mask.

Primary Threshold = 195

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

10 of 21

Methodology – Image Processing Method

10

Step 3: Perform “Connected Component Analysis” to label the grouped pixels which make up each particle.

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

1

1

1

1

0

0

0

0

0

1

1

1

1

1

1

0

0

0

1

1

1

1

1

1

1

1

0

0

1

1

1

1

1

1

1

1

0

0

1

1

1

1

1

1

1

1

0

0

0

1

1

1

1

1

1

0

0

0

0

0

1

1

1

1

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

1

1

1

1

0

0

0

0

0

1

1

1

1

1

1

0

0

0

1

1

1

1

1

1

1

1

0

0

1

1

1

1

1

1

1

1

0

0

1

1

1

1

1

1

1

1

0

0

0

1

1

1

1

1

1

0

0

0

0

0

1

1

1

1

0

0

0

0

0

0

0

0

0

0

0

0

0

…

Particle 20

Particle 21

Particle 22

…

Particle n

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

11 of 21

Methodology – Image Processing Method

11

Due to light scattering, particles appear larger on image than they actually are.

​

Step 4: Further thresholding within each labelled particle.

Thresholding’s

Considered as

Confidence level / Factor

Grayscale > 195

Background

0%

​

​

​

​

​

​

Each particle’s pixel size (in 2d) = number of “edge area” pixels x 50% + number of “core area” pixels x 100%

Step 5: Each particle’s measured size (in 2d) = its pixel size x width per pixel (~24µm per pixel)

and then, particle diameter, surface area, and volume.

165 < Grayscale < 195 “Edge area” of particle 50%

Grayscale < 165 “Core area” of particle 100%

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

12 of 21

To Test Algorithms

Testing samples are clean water, dosed with artificial particles with different standard size ranges: 38-45µm, 53-63µm, 75-90µm, 125-150µm, and 300-355µm.

12

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

13 of 21

Results

13

Volume-Weighted Particle Size Distribution

Particle Diameter (µm)

Density Distribution (% in volume)

Cumulative Distribution

125µm

150µm

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

14 of 21

Results

14

D50 = 50% of particles in sample are smaller than this diameter

Of course, they are weighted by volume.

Sensor-measured D50 particle size (µm)

Known D50/Midpoint of dosed particle size range (µm)

Volume-Weighted D50 Comparison

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

15 of 21

The selected camera for the sensor…

BoSLcam – a low-cost camera sensor:

  • 15mm x 70mm miniature size, powered by 3.3VDC, cost less than €50.
  • OV7675 camera sensor, VGA 640x480 resolution (0.3MP).
  • SD card for local image storage, SIM card for uploading image to FTP server.

15

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

16 of 21

The low-cost camera for the sensor…

Issue 1: lower camera resolution than smart phone camera…

Solution: 25x magnification macro lens -> 30µm/pixel in theory

​

16

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

17 of 21

The low-cost camera for the sensor…

Issue 2: more image noise than smart phone image…

Solution: de-noise and sharpen grayscale image using Fourier transform and high-pass filters.

​

17

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

18 of 21

Results

18

Volume-Weighted Particle Size Distribution

Particle Diameter (µm)

Density Distribution (% in volume)

Cumulative Distribution

75µm

90µm

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

19 of 21

Volume-Weighted D50 Comparison Result

19

Sensor-measured D50 particle size (µm)

Known D50/Midpoint of dosed particle size range (µm)

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

20 of 21

Future Work

  • Fine tune image processing algorithm for low-cost camera and remove subjectiveness of thresholding method;
  • Test with real environmental sample and compare with particle size analyser;
  • Make a sensor that is deployable in the field and field-deploy the sensor to compare results with manually collected and analysed sample;
  • Explore potential of measuring turbidity and TSS using the same images.

20

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology

21 of 21

Thank you for your attention!

21

bosl.com.au

Contact me: Canwei.Pang@monash.edu

Low-Cost 3-in-1 Sensor

(water depth, EC, temp)

Low-Cost Optical Turbidity Sensor

Low-Cost Out-of-water Radar Depth and Velocity Sensor

Miao Wang

Session 2.6a

10AM Friday

ICUD 2024 - June 2024 – Delft

16th International Conference on Urban Drainage 2024

Monash University & Queensland University of Technology