LoRFT: Benchmarking Long-Range Vehicle Trajectory Reconstruction from Fixed Highway Cameras

2026-07-23 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Data Science & Analytics · Depth: Expert, extended

Summary

LoRFT is introduced as the first open benchmark for long-range vehicle trajectory reconstruction from fixed highway cameras, addressing the challenge of tracklet fragmentation in distant road regions due to perspective compression and scale decay. This benchmark comprises 22 expressway surveillance scenes, 366,109 video frames, 6,601 manually verified trajectories, and 2,694,889 bounding boxes, pairing reliable near-field tracklets with context-verified far-range references. Alongside LoRFT, Map-RSTNet, a map-aware residual Seq2Seq model, is proposed. Map-RSTNet reconstructs distant trajectories by anchoring sequence modeling in a road-geometry-aligned state space and dynamically refreshing local road geometry during decoding. Experiments show Map-RSTNet reduces baseline errors on LoRFT by 11.0% on ADE, 15.4% on FDE, and 10.5% on 5-second RMSE. The dataset, evaluation scripts, and source code are publicly available.

Key takeaway

For Computer Vision Engineers developing long-range vehicle tracking systems from fixed highway cameras, you should integrate road-geometry-aware modeling. This approach significantly reduces errors in reconstructing distant vehicle trajectories where automatic tracking fails. Consider adopting techniques like dynamic geometry refresh and road-aligned state encoding to maintain accuracy over extended horizons, enabling more robust traffic analysis and autonomous driving evaluations.

Key insights

Fixed highway cameras can extend usable trajectory records by reconstructing fragmented far-range tracklets using road-geometry-aware modeling.

Principles

• Perspective compression degrades far-range tracking.
• Road geometry improves trajectory reconstruction.
• Dynamic geometry refresh limits error accumulation.

Method

Map-RSTNet converts image-space tracklets to a road-aligned state, aligns bidirectional traffic, and reconstructs distant segments via autoregressive residual decoding with dynamic geometry refresh.

In practice

• Use road-aligned state encoding for fixed-camera views.
• Implement local offset encoding for stable motion prediction.
• Dynamically refresh geometry features during long-range decoding.

Topics

• Vehicle Trajectory Reconstruction
• Fixed Highway Cameras
• LoRFT Benchmark
• Map-RSTNet
• Road Geometry Modeling
• Spatio-Temporal Data Mining

Code references

• YvfanZhu/LoRFT

Best for: AI Scientist, Computer Vision Engineer, Research Scientist

Related on AIssential

• LoRFT: Benchmarking Long-Range Vehicle Trajectory Reconstruction from Fixed Highway Cameras — The LoRFT benchmark and Map-RSTNet model advance long-range vehicle trajectory reconstruction using road-geometry awareness.
• RECO: Region-Aware Compensation for Extrinsic Perturbations in Roadside 3D Detection — RECO improves roadside 3D detection by compensating for camera extrinsic perturbations using region-aware pose offsets and a novel reprojection loss.
• TRIG: Trajectory-Rig Decoupled Metric Geometry Learning — Decoupling ego-trajectory and camera-rig components in multi-camera systems improves metric geometry and pose estimation for autonomous driving.
• Geometry-Aware Infrastructure-Anchored Denoiser for UWB Sensing and Work-Zone Reconstruction — Geometry-aware range denoising significantly improves UWB-based work-zone reconstruction accuracy by coupling temporal modeling with spatial consistency.
• Video Reconstruction using Diffusion-based Image-to-Video Generation with Trajectory Guidance — Trajectory-guided diffusion models can reconstruct missing video frames by integrating external sensor data.
• Understanding Identity Continuity in Thermal Video through Scene-Level Consistency — High-precision trajectory relinking using scene-level consistency effectively recovers identity in thermal video MOT.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.