Sidewalk Hazard Detection Using a Variational Autoencoder and One-Class SVM

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Outdoor navigation presents safety challenges due to unpredictable hazards.
• Effective hazard detection is critical for safe navigation systems.

Purpose of the Study:

• To develop a low-cost, portable sidewalk hazard detection system.
• To combine Variational Autoencoder (VAE) and One-Class Support Vector Machine (OCSVM) for robust anomaly identification.

Main Methods:

• Utilized a wearable RGB camera for data acquisition.
• Trained a VAE on normal sidewalk data to learn appearance patterns.
• Employed OCSVM to classify VAE-identified anomalies as hazardous or non-hazardous.

Main Results:

• Achieved an Area Under the Curve (AUC) of 0.92 and an F1 score of 0.85.
• Outperformed baseline anomaly detection models in outdoor sidewalk navigation tasks.
• Demonstrated robust performance on a custom dataset of over 28,000 images.

Conclusions:

• The proposed VAE + OCSVM hybrid method provides a reliable solution for real-world sidewalk hazard detection.
• The system offers a practical approach for enhancing the safety of outdoor navigation.
• This research contributes to the advancement of autonomous navigation and computer vision applications.