An evaluation of machine learning methods for speed-bump detection on a GoPro dataset

Area of Science:

• Computer Science
• Robotics
• Geospatial Analysis

Background:

• High-resolution road maps are crucial for autonomous vehicles, especially when sensors fail in adverse weather.
• Current mapping methods struggle with real-time updates and detecting vertical road anomalies.

Purpose of the Study:

• To develop an automated methodology for mapping road anomalies, specifically speed bumps.
• To assess the effectiveness of Machine Learning (ML) algorithms using off-the-shelf camera data.

Main Methods:

• Data acquisition using a GoPro camera across various speed bump types.
• Application and comparison of three ML classification techniques: Naive Bayes, Multi-Layer Perceptron, and Random Forest (RF).

Main Results:

• The Random Forest algorithm demonstrated high classification accuracy exceeding 96% in identifying speed bumps.
• The proposed methodology successfully automated the detection of vertical road anomalies from camera data.

Conclusions:

• The developed ML-based approach offers a fast and accurate method for creating detailed maps of vertical road anomalies.
• This technique has significant potential for integration into surveying vehicles for real-time road mapping updates.