Detection of Road Potholes by Applying Convolutional Neural Network Method Based on Road Vibration Data

Area of Science:

• Road transportation safety
• Vibration analysis
• Machine learning for infrastructure monitoring

Background:

• Potholes pose significant risks to driving comfort, vehicle maintenance costs, and road safety.
• Existing detection methods often require specialized equipment, increasing costs for vehicle owners.

Purpose of the Study:

• To develop a cost-effective pothole detection system utilizing readily available smartphone technology.
• To introduce a novel application of Convolutional Neural Networks (CNNs) for analyzing road surface vibrations.

Main Methods:

• An iOS application was developed to collect vibration data (accelerometer, gyroscope) and GPS coordinates.
• Road vibration data were converted into visual representations (pixel-based images).
• Multiple CNN models with varying architectures were trained and validated for pothole identification.

Main Results:

• CNN models demonstrated high accuracy (93.24%) and low loss (0.2948) during initial validation.
• A two-stage validation process confirmed the system's effectiveness in real-world conditions.
• Field studies showed successful pothole detection rates between 80% and 87%.

Conclusions:

• Smartphone-integrated vibration sensing combined with CNNs offers a viable and accurate solution for pothole detection.
• This approach reduces the need for expensive dedicated hardware, making road monitoring more accessible.
• The developed system contributes to improved road safety and reduced transportation costs.