Abnormal Driving Pattern Detection from GPS Trajectories Using Vision Transformer
DOI: 10.21203/rs.3.rs-8653475/v1
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This paper addresses the challenge of detecting abnormal driving patterns from GPS trajectories, a task critical for enhancing road safety and identifying early signs of Mild Cognitive Impairment (MCI) in older adults. The authors identify a significant limitation in existing deep learning approaches: GPS trajectories vary widely in length, shape, and route, making it difficult to standardize inputs for fixed-size models without losing contextual information. Traditional methods often segment trips or restrict analysis to identical origins and destinations, which is impractical for real-world data. To overcome this, the study proposes a novel framework that converts raw GPS coordinates into fixed-size binary grid images, preserving the complete spatial structure of each trip regardless of its duration or complexity. The methodology involves transforming each driving trip into a 128×128 binary matrix, where cells are marked as visited or unvisited based on the trajectory’s bounding box. This “trajectory-to-image” encoding allows the application of Vision Transformers (ViT), which leverage self-attention mechanisms to capture both local and global spatial dependencies. The model architecture processes these grids by partitioning them into patches, embedding them, and passing them through six transformer encoder layers to classify trips as normal or abnormal. To address class imbalance, the authors applied rotational data augmentation (90°, 180°, 270°) to the minority class. The experiments were conducted on a large-scale dataset comprising over 72 million trajectory records from older adult drivers (aged 65+), labeled by clinicians using comprehensive neuropsychological evaluations. The ViT model was compared against Convolutional Neural Network (CNN) and ResNet baselines. The results demonstrate that the proposed ViT-based framework significantly outperforms traditional models, achieving an F1 score of 94%. The study found that increasing the dataset size from four months to two years consistently improved performance across all models. Furthermore, rotational data augmentation enhanced model robustness, with the ViT showing the most pronounced improvement, indicating its superior ability to capture rotation-invariant spatial dependencies. The binary grid representation proved effective in encoding interpretable spatial patterns, such as cyclic routes and inefficient navigation, which are indicative of cognitive decline. The significance of this work lies in its contribution to non-invasive, privacy-preserving monitoring of cognitive health. By effectively modeling longitudinal driving behavior, the approach enables earlier identification of at-risk individuals, supporting timely clinical interventions and safer transportation systems. The study validates that transformer-based architectures, when combined with spatial grid representations, offer a scalable solution for analyzing heterogeneous spatiotemporal data, bridging the gap between telematics and public health monitoring.
Key finding
A Vision Transformer model applied to binary grid representations of GPS trajectories achieved an F1 score of 94% in detecting abnormal driving patterns, significantly outperforming baseline CNN and ResNet models.
Methodology
dataset
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via author_sweep_intake on 2026-05-28.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-28 |
| archive | success | canonical_url | — | — | 1 | 2026-06-04 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-04 |
| chunk | success | chunk | — | — | 1 | 2026-06-04 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-04 |
| enrich | success | crossref | — | — | 2 | 2026-06-04 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 2 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 15 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Theoretical Contribution: computational model