A deep learning approach for transportation mode identification using a transformation of GPS trajectory data features into an image representation

• DOI: 10.1007/s41060-024-00510-3
• Publication Date: 2024-02-12T13:02:06Z
• Authors: Ricardo Ribeiro; Alina Trifan; António J. R. Neves
• Keywords: Representation; Identification; Mode (computer interface)

Abstract Global positioning system data play a crucial role in comprehending an individual’s life due to its ability provide geographic positions and timestamps. However, it is challenge identify the transportation mode used during trajectory large amount of spatiotemporal generated, distinct spatial characteristics exhibited. This paper introduces novel approach for identification by transforming features into image representations employing these images train neural network based on vision transformers architectures. Existing approaches require predefined temporal intervals or sizes, limiting their adaptability real-world scenarios characterized several lengths inconsistent intervals. The proposed avoids segmenting changing trajectories directly extracts from data. By mapping pixel location generated using dimensionality reduction technique, are created deep learning model predict five transport modes. Experimental results demonstrate state-of-the-art accuracy 92.96% Microsoft GeoLife dataset. Additionally, comparative analysis was performed traditional machine method offers accurate reliable applicable scenarios, facilitating understanding mobility.