Road Damage Detector Based on Vibration and Trajectory Using CNN-LSTM Algorithm

The increasing volume of traffic due to social activities has led to road damage, such as potholes, which pose risks of accidents, vehicle damage, and discomfort. Manual road condition inspections are time-consuming and costly, necessitating an efficient automated solution. This study aims to develop a road damage detection system based on vibrations from an accelerometer and gyroscope sensor (MPU6050), GPS module, and camera to identify road types (normal, pothole, speed bump, speed trap) in real-time. The system employs a CNN-LSTM (Convolutional Neural Network-Long Short-Term Memory) algorithm, where CNN extracts vibration features and LSTM analyzes temporal patterns, embedded in a ESP32-S3 microcontroller. The location of road damage is recorded via GPS, uploaded to a website, and documented with photos. Testing shows a prediction accuracy of 92%–100% across all road type classifications. The system successfully detects potholes automatically without misclassifying them as other road types. This system offers a fast, cost-effective, and accurate solution for road condition monitoring, with potential integration into infrastructure management systems.