An Automated Machine-Learning Approach for Road Pothole Detection Using Smartphone Sensor Data.

Area of Science:

• Civil Engineering
• Transportation Engineering
• Computer Science

Background:

• Road monitoring is crucial for safety and infrastructure preservation.
• Traditional methods using instrumented vehicles are costly and limited in scope.
• Smartphones offer a ubiquitous and accessible platform for data collection.

Purpose of the Study:

• To develop an automatic pothole detection system using smartphone sensors.
• To evaluate the effectiveness of machine learning classifiers for pothole identification.
• To assess the universality and robustness of the proposed method across different road types.

Main Methods:

• Collected road condition data using smartphones and dedicated vehicles.
• Applied signal processing to extract time and frequency domain features from vibration data.
• Utilized machine learning classifiers, including Random Forest, for pothole classification.

Main Results:

• Time and frequency domain features were superior for pothole detection.
• The Random Forest classifier achieved the highest performance: 88.5% precision and 75% recall.
• The method demonstrated robustness and universality across various road conditions.

Conclusions:

• Smartphone-based vibration sensing provides a viable alternative for road surface monitoring.
• The proposed system offers a cost-effective and scalable solution for pothole detection.
• Further validation can enhance road maintenance strategies and infrastructure management.