Towards Enhancing Traffic Sign Recognition through Sliding Windows.

Area of Science:

• Computer Vision
• Machine Learning
• Autonomous Systems

Background:

• Automatic Traffic Sign Detection and Recognition (TSDR) is crucial for autonomous driving safety.
• Existing Traffic Sign Recognition (TSR) methods using Machine Learning (ML) lack perfect classification capabilities.
• Misclassification of traffic signs poses severe risks to safety, environment, and infrastructure.

Purpose of the Study:

• To develop a reliable TSR system using ML-based classifiers.
• To improve the accuracy and reduce misclassifications in traffic sign recognition.
• To leverage sliding window frame analysis with advanced ML techniques.

Main Methods:

• A TSR system was developed analyzing a sliding window of sensor-sampled frames.
• The system employed Long Short-Term Memory (LSTM) networks and Stacking Meta-Learners.
• Stacking Meta-Learners were used to combine base-learning classification episodes for improved meta-level classification.

Main Results:

• The proposed Stacking Meta-Learners approach dramatically reduced sign misclassifications.
• Perfect classification was achieved on all three considered publicly available datasets.
• The sliding window approach demonstrated significant potential for efficient TSR.

Conclusions:

• The novel ML-based TSR system effectively enhances road safety for autonomous driving.
• Stacking Meta-Learners offer a robust solution for accurate traffic sign recognition.
• The sliding window technique combined with advanced ML shows promise for real-world applications.