Driving Environment Inference from POI of Navigation Map: Fuzzy Logic and Machine Learning Approaches.

Area of Science:

• Intelligent Transportation Systems
• Machine Learning for Environmental Perception

Background:

• Predictive vehicle control requires understanding the driving environment.
• Existing methods may lack efficiency in real-time context inference.

Purpose of the Study:

• To efficiently infer five key driving environments: shopping, tourist, public station, motor service, and security zones.
• To develop a robust inference framework using Point of Interest (POI) data.
• To evaluate different inference systems for driving environment classification.

Main Methods:

• Utilized Point of Interest (POI) data from navigation maps as semantic clues.
• Framed the inference task as a multilabel classification problem.
• Developed a statistical approach for numerical POI feature extraction.
• Investigated Fuzzy Inference System (FIS), Support Vector Machine (SVM), and Multilayer Perceptron (MLP) inference engines.
• Implemented 11 inference engine variants using independent and unified strategies.

Main Results:

• The proposed framework demonstrated good generalization across different inference systems.
• The MLP-based inference engine with a unified strategy achieved the highest overall F1 score of 0.8699.
• The best performing model exhibited an exceptionally fast inference time of 0.0002 milliseconds per sample.

Conclusions:

• The developed inference framework is both generalizable and efficient for driving environment recognition.
• The MLP model with a unified strategy offers a promising solution for real-time applications in autonomous driving.
• POI data combined with advanced machine learning effectively enhances situational awareness for vehicles.