Using Sensor Fusion and Machine Learning to Distinguish Pedestrians in Artificial Intelligence-Enhanced Crosswalks

• Published in: Electronics (Basel) 2023-12, Vol.12 (23), p.4718
• Main Authors: Lozano Domínguez, José Manuel, Redondo González, Manuel Joaquín, Davila Martin, Jose Miguel, Mateo Sanguino, Tomás de J.
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Artificial intelligence; Automation; Fuzzy logic; Hypotheses; Image processing; Infrastructure; Lighting; Machine learning; Magnetic fields; Pattern recognition; Pedestrian crossings; Pedestrian safety; Pedestrians; Radar systems; Radio; Reinforcing fibers; Roads; Roads & highways; Sensors; Signalling systems; Support vector machines; Traffic accidents & safety; Traffic safety; Vehicles
• ISSN: 2079-9292; 2079-9292
• DOI: 10.3390/electronics12234718

Pedestrian safety is a major concern in urban areas, and crosswalks are one of the most critical locations where accidents can occur. This research introduces an intelligent crosswalk, employing sensor fusion and machine learning techniques to distinguish the presence of pedestrians and drivers. Upon detecting a pedestrian, the system proactively activates a warning light signal. This approach aims to quickly alert nearby people and mitigate potential dangers, thereby strengthening pedestrian safety. The system integrates data from radio detection and ranging sensors and a magnetic field sensor, using a hierarchical classifier. The One-Class support vector machine algorithm is used to classify objects in the radio detection and ranging data, while fuzzy logic is used to filter out targets from the magnetic field sensor. Additionally, this work presents a novel method for the manufacture of the road signaling system, using mixtures of resins, aggregates, and reinforcing fibers that are cold-injected into an aluminum mold. The mechanical, optical, and electrical characteristics were subjected to standardized tests, validating its autonomous operation in real-world conditions. The results revealed the system’s effectiveness in detecting pedestrians with a 99.11% accuracy and a 0.0% false-positive rate, marking a substantial improvement over the previous fuzzy logic-based system with an 81.33% accuracy. Attitude testing revealed a significant 33.33% reduction in pedestrian erratic behavior and a substantial decrease in driver speed (32.83% during the day and 70.6% during the night) compared to conventional crossings. Consequently, this comprehensive work offers a unique solution to pedestrian safety at crosswalks by showcasing the potential of machine learning techniques, particularly the One-Class support vector machine algorithm, in advancing road safety through precise and reliable pattern recognition.