 | 2025 13-th International Scientific Conference COMPUTER SCIENCE |  |
|---|
Invited paper "Features Extraction for Vehicle Re-Identification Using Magnetic Sensing" by Prof. Darius Andriukaitis,
Kaunas University of Technology, Kaunas, Lithuania
Abstract: Road traffic is growing at an unprecedented pace, with over 1.5 billion vehicles in use worldwide. This creates significant challenges, traffic congestion, longer travel times, higher fuel consumption, and increased air pollution, especially in cities centres. The transportation sector is a major source of greenhouse gas emissions, accounting for about a quarter of emissions in the European Union. To address these issues, Intelligent Transportation Systems (ITS) combine electronics, information processing, and communication technologies to make transportation safer, more efficient, and more sustainable.
A key function of ITS subsystem is vehicle re-identification, where accurate determination of a vehicle in two different places is crucial. This capability enables better traffic monitoring, congestion management, and route optimization. However, many existing systems have technical limitations. Our research group “Interactive electronic systems” work presents a new, depersonalized re-identification approach based on magnetic sensing. When a vehicle passes over a magnetic sensor array, it generates a unique magnetic signature, independent of travel direction or sensor orientation.
To ensure data quality the collected magnetic signatures were from two different systems, focusing on vehicles recorded multiple times. The experiments were conducted with different length subsets using artificial neutral network and Recurall neural network trained for regression problem where ANN for regression is the most suitable for valid sensor selection. A new method was developed to identify the most relevant sensors in the array, achieving more than 94% accuracy.
In order to make signatures suitable for similarity evaluation, an algorithm for signature preprocessing was developed, which includes filtering for noise reduction, resampling to a fixed number of points and various useful feature extraction using various distance measures, signal compression and time-frequency analysis.
The investigation of signals features extraction methods (Euclidean difference, signal divided in regions asymmetry calculation, compression through zones and polynomial fitting, dynamic time warping, Pearson correlation coefficient, 2D-correlation) led to the features determination that have minimal overlapping between the two classes. Euclidean distance overlapping does not exceed 14%, Pearson correlation coefficient 7%. These features are strong candidates for distinguishing between the same and different vehicles.
These results show that magnetic sensing can provide a reliable, privacy-preserving method for vehicle re-identification, supporting the development of smarter and greener transportation systems.
Invited paper "Application of COMSOL in the cable ampacity calculations" by Prof. Dardan Klimenta, University of Priština in Kosovska Mitrovica, Serbia.
Last changed on 13.09.2025, 16:30:27