E-Scooter Mobility Sensing for Smart City Public Safety and Asset Management

By Ethan Thomas
Slide 1: Title slide presenting E-Scooter Mobility Sensing for Smart City Public Safety and Asset Management by Ethan Thomas

Slide-1

Presenter: Ethan Thomas, Computer Science Department, Columbia University School of Engineering and Applied Sciences

REU Mentor: Jinwoo Jang, Ph.D., Florida Atlantic University

Slide 2: Background information about road surface conditions and their impact on safety

Slide-2

Background

  • Incorporates the Center for Smart Streetscapes' Situational Awareness thrust.
  • Roads and sidewalks are essential for connecting people.
  • Bumpy riding surfaces are unsafe, especially at high speeds and in adverse weather conditions.
  • Smooth surfaces reduce accident risk for wheelchairs, strollers, scooters, etc.
  • How do we quantify and classify these surfaces?
Slide 3: Research objectives for using accelerometers to classify riding surface types

Slide-3

Objective

  • Use accelerometers on e-scooters to classify riding surface types.
  • Applications include detection of surfaces for unsafe ride prevention.
  • Riding surfaces in poor conditions can be reported for maintenance.
Slide 4: System design showing hardware and software components used in the research

Slide-4

System Design

Hardware

  • Xiaomi Scooter
  • Camera (Logitech C920)
  • Raspberry Pi
  • IMU (Accelerometer)
  • GPS

Software

  • LCM (Lightweight Communication Protocol)
  • BLE (Bluetooth Low Energy)
  • Phidget22 Library
  • Raspbian OS/Systemd/OpenCV
Slide 5: Data stream visualization showing accelerometer readings for different surface types

Slide-5

Data Collection

Graph showing accelerometer data for different surface types:

Collected accelerometer data on three surface types.

  • Concrete w/ Expansion Joint
  • Asphalt
  • Pavers
  • Stopped

Data Stream

Acceleration

0.19646, 0.04279, 1.06407

GPS

26.372973, -80.100547, 10.9

Speed

8.83

Slide 6: Data processing methodology showing segmentation and labeling approach

Slide-6

Data Processing

  • Split data into segments of 128 data points (around 10 seconds each).
  • Labeled each segment as one of the four surface types using the video reference.
  • Used each segment/label pair as the input/output for training the Machine Learning model.

Graph shows Time vs Acceleration with z-axis (down) and x-axis (forward) and y-axis (right) measurements.

Slide 7: Machine learning approaches and results comparing LSTM and FFT methods

Slide-7

Machine Learning

  • Implemented a LSTM Recurrent Neural Network using PyTorch, with poor (~50% accuracy) results.
  • Trained a model using Fast Fourier transform and tsfresh feature extraction with 80% accuracy.
Slide 8: Research conclusions summarizing key achievements and deliverables

Slide-8

Conclusions

  • Prepared e-scooter for data collection using IoT devices.
  • Collected over three hours of data on different surface types.
  • Applied Machine Learning models to the collected data.
Slide 9: Future work outlining next steps and improvements for the research

Slide-9

Future Work

  • More data collection on more varied surfaces.
  • Put GPS data, and potentially speed data, to use.
  • Fine-tune machine learning model for greater accuracy.
  • Equip the Raspberry Pi to run the model in real-time.
Slide 10: Questions and feedback session invitation

Slide-10

Questions & Feedback

Last slide: Contains plain text stating 'End of presentation. Click the right arrow to return to beginning of slide show.'

End of Presentation

Click the right arrow to return to the beginning of the slide show.

For a downloadable version of this presentation, email: I-SENSE@FAU.

Additional Information
The Institute for Sensing and Embedded Network Systems Engineering (I-SENSE) was established in early 2015 to coordinate university-wide activities in the Sensing and Smart Systems pillar of FAU’s Strategic Plan for the Race to Excellence.
Address
Florida Atlantic University
777 Glades Road
Boca Raton, FL 33431
i-sense@fau.edu