Yuqian Chen
Yuqian Chen has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
Bing Li
1 shared publication- Jingfeng Ma
Southeast University
1 shared publication - Juyuan Yin
Faculty of Transportation Engineering, Kunming University of Science and Technology, Kunming, China
1 shared publication - Xuejian Chen
Key Laboratory of Road and Traffic Engineering of the State Ministry of Education, Tongji University, Shanghai, China
1 shared publication
Latest Publications
- Open Access
- 5 June 2026
- Published by Elsevier in SSRN Electronic Journal
CITR-GAN: Collective Interaction-Aware Trajectory Reconstruction for Sparse Detector Environments on Freeways
Millimeter-wave radar and video integrated detectors have been widely used in intelligent transportation systems for vehicle trajectory data collection. However, to balance coverage and cost, these detectors are often deployed sparsely, resulting in a considerable amount of missing trajectory data. This approach poses considerable challenges for detection, making continuous vehicle tracking in complex traffic scenarios extremely difficult. To overcome this challenge, we employ deep generative learning to reconstruct vehicle trajectories. Specifically, we propose CITR-GAN, a collective interaction-aware trajectory reconstruction method, which integrates vehicle interaction features with input-conditioned and conditional GAN models to achieve accurate and reliable trajectory reconstruction. By learning latent trajectory distributions, the proposed method enables efficient inference and achieves accurate longitudinal trajectory reconstruction on the real-world HighD dataset. Its effectiveness is further validated on the NGSIM dataset, where more congested traffic conditions, dense vehicle interactions, and complex longitudinal motion patterns pose greater reconstruction challenges. Under the challenging long blind-zone setting of 400 m, CITR-GAN achieves 2.43 m MAE, 3.69 m RMSE, and 1.60% MAPE, with a 19.0% MAE reduction over the best baseline method. Assuming a representative detector range of 250 m per site, it can potentially reduce detector density by up to 44.4% while maintaining blind-zone reconstruction capability. Results demonstrate that CITR-GAN not only improves reconstruction accuracy under challenging long blind-zone conditions but also provides a practical solution for enhancing sensing continuity with lower infrastructure requirements.
- 5 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessCITR-GAN: Collective Interaction-Aware Trajectory Reconstruction for Sparse Detector Environments on Freeways
Millimeter-wave radar and video integrated detectors have been widely used in intelligent transportation systems for vehicle trajectory data collection. However, to balance coverage and cost, these detectors are often deployed sparsely, resulting in a considerable amount of missing trajectory data. This approach poses considerable challenges for detection, making continuous vehicle tracking in complex traffic scenarios extremely difficult. To overcome this challenge, we employ deep generative learning to reconstruct vehicle trajectories. Specifically, we propose CITR-GAN, a collective interaction-aware trajectory reconstruction method, which integrates vehicle interaction features with input-conditioned and conditional GAN models to achieve accurate and reliable trajectory reconstruction. By learning latent trajectory distributions, the proposed method enables efficient inference and achieves accurate longitudinal trajectory reconstruction on the real-world HighD dataset. Its effectiveness is further validated on the NGSIM dataset, where more congested traffic conditions, dense vehicle interactions, and complex longitudinal motion patterns pose greater reconstruction challenges. Under the challenging long blind-zone setting of 400 m, CITR-GAN achieves 2.43 m MAE, 3.69 m RMSE, and 1.60% MAPE, with a 19.0% MAE reduction over the best baseline method. Assuming a representative detector range of 250 m per site, it can potentially reduce detector density by up to 44.4% while maintaining blind-zone reconstruction capability. Results demonstrate that CITR-GAN not only improves reconstruction accuracy under challenging long blind-zone conditions but also provides a practical solution for enhancing sensing continuity with lower infrastructure requirements.