- The classification model is still in development phase, furthermore, the publication of additional algorithms (like map-matching and map-display solutions) can be expected in the near future too.
Poor-quality roads pose a significant safety risk, it can reduce the lifetime of mechanical components, furthermore the reconstruction costs increase exponentially as the surface deteriorates. It is therefore important to repair road defects quickly, which requires knowledge of the condition of the road network.
However, the current available systems has complex hardware, increasing the operating costs. Thus, these systems are mainly used on higher ranked roads.
Development of a universally applicable, cost-effective and easy-to-use system for measuring and classifying road sections according to their quality. The target group of the development is the road management companies, for whom the new type of device means a financial and time return.
Main contributions:
- Development of the complete instrumentation unit with vibration sensors placed in vehicle cabin
- Data warehouse implementation with Microsoft Azure
- Data processing and ML model development in Python:
- Data preprocessing and cleaning
- EDA
- Machine-in-the-loop data labelling process with unsupervised-learning model
- Supervised machine learning model development
- GPS coordinate reverse-geocoding and route map-matching
- Interactive visualization
- Pipelining
Dataset:
- First-party data collected with custom-developed intrument
We developed a system based on a new approach that can classify road sections according to quality based on minimal sensor and machine learning.
-
Technical contributions:
- Complete end-to-end solution
- Industry-standard implementation with Python and Microsoft Azure data warehouse
- Machine-in-the-loop annotation process for data labelling and machine-learning based pavement classification algorithm
- Geocoded and map-matched road quality results, making the search easier in the database
- Online interactive visualization with Folium
-
Business-related contributions:
- Road network monitoring costs for road management companies can be significantly reduced
- Measurements can be more time efficient and faster
- The low cost of installation makes the solution highly scalable
- Based on a comprehensive database, the necessary and preventive road rehabilitation works can be optimised
- As a result, road safety has increased significantly
Methods Used
- Data Cleaning
- Data Exploration
- Data Visualization
- Machine-in-the-loop labelling
- Machine learning development and deployment, MLops
- Pipelining
Tools
- Python
- SQL
- Microsoft Azure
Coming soon.