Road Surface Recognition Based on DeepSense Neural Network using Accelerometer Data

This work is presented and accepted for publication in IV 2020.

In this paper, we proposed the methodology of recognizing "smooth", "bumpy", and "rough" road surfaces by using a unified sensor fusion framework DeepSense [1]. The raw accelerometer data is collected from a mobile phone, and processed on a desktop afterwards. During the preprocessing stage, raw sensory data is divided into small chunks and preprocessed using tsfresh [2] to extract time-series features. We implemented the DeepSense deep learning framework for the first time and succeeded in deploying it for road surface recognition on the desktop. One can also make it possible on a real-time on-board application.

Motivation

We are releasing our approach's source code for Road Surface Recognition based on accelerometer sensor to share with the scientific community and industry with the aim of the collaboration with people around the world to push the boundaries in the field of intelligent transportation systems.

Experiment results

Dataset

The collected accelerometer data can be found here. Each file collected is a session and stored in csv format with the following fields:

Field Name	Data Type	Description
Timestamp	Int	recorded in milliseconds, starts from 00:00:00.0, 1st of January, 1970 UTC
X-raw	Float	Raw accelerometer data for X axis
Y-raw	Float	Raw accelerometer data for Y axis
Z-raw	Float	Raw accelerometer data for Z axis
X-axis	Float	Virtually-oriented accelerometer data for X axis. This accelerometer data is virtually transformed from the phone’s coordinate frame to the world coordinate frame
X-axis	Float	Virtually-oriented accelerometer data for Y axis
X-axis	Float	Virtually-oriented accelerometer data for Z axis
Road_type	String	One of the "Smooth", "Bumpy" or "Rough" road type

Deployment

For a list of all source code dependencies, please see here.

1. Clone the repo to your local drive.
2. Setup your own Python environment and install the requirements.txt by command pip3 install -r requirements.txt.
3. Due to the inconsistency of installing the tensorflow, you may need to install it manually.

Usage

We organized the codes in a modular way to make it convenient for customized usages. Before start, make sure the folders in the root directory is organized in the following way:

• Data folder at ./data/
• Execution root at ./src/

In the main.py, there's steps to run cross validation.

cd src/
python main.py

Licence

Our model is released under a GPLv3.0 license.

For a closed-source version of the source code for commercial purposes, please contact the authors: Wu and Hadachi

Contributors

Shan Wu; Amnir Hadachi.

Citation

If you use our implementation or our dataset in an academic work, please cite:

@inproceedings{Shan2020Road,
  title={Road Surface Recognition Based on DeepSense Neural Network using Accelerometer Data},
  author={Wu, Shan and Hadachi, Amnir},
  booktitle={Proceedings of the IEEE Intelligent Vehicles Symposium 2020},
  year={2020},
  organization={IEEE}
}

Publicated in IEEE Explore. Preprint version of the paper is here.