pith. sign in

arxiv: 2411.02349 · v2 · submitted 2024-11-04 · 📡 eess.IV

Drone Data Analytics for Measuring Traffic Metrics at Intersections in High-Density Areas

Qingwen Pu , Yuan Zhu , Junqing Wang , Hong Yang , Kun Xie , Shunlai Cui This is my paper

Pith reviewed 2026-05-23 17:25 UTC · model grok-4.3

classification 📡 eess.IV
keywords drone videoUAV trajectorieshigh-density intersectionsroad user trackinglane change analysistraffic conflictspublic datasettraffic metrics
0
0 comments X

The pith

Over 100 hours of drone video from eight intersections produces the largest public dataset of trajectories for more than one million road users in high-density urban traffic.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes a large-scale public dataset drawn from high-altitude drone videos at eight intersections in Hohhot. An enhanced YOLOUAV model and automated calibration algorithm track more than 200 vehicles per frame across more than 1 million road users and record over 50,000 complete lane changes. The resulting trajectories support generation of spatial-temporal heatmaps and identification of traffic conflicts through lane-change counts and surrogate safety measures. A sympathetic reader would care because the resource supplies extensive, ready-to-use data for traffic engineering studies in dense areas while minimizing manual processing.

Core claim

The High-Density Intersection Dataset is the largest publicly available collection of road user trajectories from high-density urban intersections. It is built from more than 100 hours of UAV video using an enhanced YOLOUAV model for target recognition and an automated calibration algorithm that produces functional data in dense flows. The dataset tracks cars, buses, and trucks, applies UAV-elevation corrections to speed and acceleration calculations, includes an offset correction step, and enables case-study analysis of intersection performance through heatmaps and conflict location.

What carries the argument

The enhanced YOLOUAV model paired with an automated calibration algorithm that converts raw UAV video into accurate multi-vehicle trajectories in high-density traffic.

If this is right

  • The dataset supplies parameters needed to evaluate intersections and overall traffic conditions.
  • Spatial-temporal data can be rendered as heatmaps of traffic flow.
  • Lane-change counts combined with surrogate measures can locate traffic conflicts.
  • UAV-elevation and offset corrections improve the accuracy of derived speed and acceleration values.
  • The methods allow simultaneous tracking of more than 200 vehicles of mixed types in dense conditions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same drone-plus-calibration pipeline could be repeated in other cities to create comparable cross-site datasets.
  • The scale of the trajectories might support training of predictive models for real-time conflict detection.
  • Public release of the data lowers the barrier for researchers without access to UAV equipment.

Load-bearing premise

The automated calibration algorithm and enhanced YOLOUAV model produce accurate trajectories and metrics in high-density flows without substantial tracking errors or the need for manual corrections.

What would settle it

Manual annotation of a random sample of video frames that shows vehicle identification or tracking error rates substantially higher than those implied by the automated output.

read the original abstract

This study employed over 100 hours of high-altitude drone video data from eight intersections in Hohhot to generate a unique and extensive dataset encompassing high-density urban road intersections in China. This research has enhanced the YOLOUAV model to enable precise target recognition on unmanned aerial vehicle (UAV) datasets. An automated calibration algorithm is presented to create a functional dataset in high-density traffic flows, which saves human and material resources. This algorithm can capture up to 200 vehicles per frame while accurately tracking over 1 million road users, including cars, buses, and trucks. Moreover, the dataset has recorded over 50,000 complete lane changes. It is the largest publicly available road user trajectories in high-density urban intersections. Furthermore, this paper updates speed and acceleration algorithms based on UAV elevation and implements a UAV offset correction algorithm. A case study demonstrates the usefulness of the proposed methods, showing essential parameters to evaluate intersections and traffic conditions in traffic engineering. The model can track more than 200 vehicles of different types simultaneously in highly dense traffic on an urban intersection in Hohhot, generating heatmaps based on spatial-temporal traffic flow data and locating traffic conflicts by conducting lane change analysis and surrogate measures. With the diverse data and high accuracy of results, this study aims to advance research and development of UAVs in transportation significantly. The High-Density Intersection Dataset is available for download at https://github.com/Qpu523/High-density-Intersection-Dataset.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript presents a dataset derived from over 100 hours of high-altitude drone video collected at eight intersections in Hohhot, China. It describes enhancements to the YOLOUAV model for target detection on UAV imagery, an automated calibration algorithm for generating trajectories in high-density traffic, updates to speed/acceleration estimation based on UAV elevation, and an offset correction method. The work claims to track more than 200 vehicles per frame, yielding trajectories for over 1 million road users (cars, buses, trucks) and over 50,000 complete lane changes; it generates spatial-temporal heatmaps and identifies conflicts via lane-change analysis and surrogate safety measures. The resulting High-Density Intersection Dataset is released publicly on GitHub.

Significance. If the tracking and calibration accuracy claims are substantiated, the work would provide a valuable large-scale public resource for traffic flow analysis, conflict detection, and UAV-based monitoring in dense urban settings, where such datasets are scarce. The public data release supports reproducibility and community extension, which strengthens the contribution.

major comments (2)
  1. [Abstract] Abstract: The central claims of 'precise target recognition,' 'accurately tracking' over 200 vehicles per frame, and reliable generation of lane-change and conflict metrics rest on the enhanced YOLOUAV plus automated calibration, yet no quantitative validation metrics (MOTA, IDF1, position RMSE, ID-switch rate, or comparison to manual ground-truth annotations on held-out dense frames) are reported. This is load-bearing for the assertion that the dataset enables trustworthy downstream analyses without substantial tracking errors.
  2. [Methods] Methods (automated calibration section): The automated calibration algorithm is presented as enabling a 'functional dataset in high-density traffic flows' that 'saves human and material resources,' but no validation results (e.g., homography error, comparison to manual calibration, or performance across density regimes) are supplied to confirm it produces accurate trajectories without manual corrections in scenes exceeding 200 vehicles per frame.
minor comments (2)
  1. [Abstract] The abstract states that speed and acceleration algorithms are 'updated based on UAV elevation' but provides no explicit equations or differences from prior UAV-based methods, reducing clarity on the technical contribution.
  2. Figure captions and table descriptions should explicitly state the number of frames or intersections used for any reported qualitative examples (e.g., heatmaps or conflict locations) to allow readers to gauge representativeness.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight important areas where additional quantitative evidence would strengthen the presentation of our dataset and methods. We address each major comment below and commit to revisions that directly respond to these points.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claims of 'precise target recognition,' 'accurately tracking' over 200 vehicles per frame, and reliable generation of lane-change and conflict metrics rest on the enhanced YOLOUAV plus automated calibration, yet no quantitative validation metrics (MOTA, IDF1, position RMSE, ID-switch rate, or comparison to manual ground-truth annotations on held-out dense frames) are reported. This is load-bearing for the assertion that the dataset enables trustworthy downstream analyses without substantial tracking errors.

    Authors: We agree that the abstract's claims regarding precision and accuracy would be more robust with explicit quantitative validation metrics. The current version emphasizes the scale of the released dataset and the public availability for community scrutiny rather than reporting standard tracking metrics. In the revised manuscript, we will add a validation subsection (likely in Methods or a new Results subsection) that reports MOTA, IDF1, position RMSE, ID-switch rates, and comparisons against manual ground-truth annotations on held-out dense frames. This addition will directly substantiate the claims and address the load-bearing concern. revision: yes

  2. Referee: [Methods] Methods (automated calibration section): The automated calibration algorithm is presented as enabling a 'functional dataset in high-density traffic flows' that 'saves human and material resources,' but no validation results (e.g., homography error, comparison to manual calibration, or performance across density regimes) are supplied to confirm it produces accurate trajectories without manual corrections in scenes exceeding 200 vehicles per frame.

    Authors: We acknowledge that the automated calibration section would benefit from quantitative validation to demonstrate its reliability in high-density conditions. The manuscript currently describes the algorithm's design and its role in enabling the large-scale dataset but does not include error metrics or comparisons. We will revise the Methods section to incorporate validation results, including homography errors versus manual calibration, performance across varying density regimes, and evidence that the method operates without manual corrections in scenes with over 200 vehicles per frame. These additions will confirm the algorithm's effectiveness as claimed. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain; empirical data processing only.

full rationale

The paper presents an empirical workflow: collection of drone video, enhancement of an existing YOLOUAV detector, application of an automated calibration routine, and extraction of trajectories/metrics from the resulting data. No equations, fitted parameters, or self-citations are invoked to derive one quantity from another by construction; counts of vehicles, lane changes, and derived heatmaps/conflict measures are direct outputs of the processing pipeline applied to external video input. The central claims rest on the scale and public release of the collected dataset rather than any self-referential reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The work rests on standard computer-vision tracking assumptions and the practical feasibility of high-altitude drone filming; no free parameters or invented entities are introduced in the abstract.

axioms (2)
  • domain assumption Standard assumptions of object-detection models such as YOLO remain valid when applied to UAV imagery of dense traffic
    The paper builds directly on the YOLOUAV model without re-deriving its performance guarantees.
  • domain assumption Drone video can be calibrated to ground-plane coordinates using elevation and offset corrections without introducing large systematic errors
    The automated calibration algorithm is presented as sufficient for high-density flows.

pith-pipeline@v0.9.0 · 5803 in / 1502 out tokens · 30694 ms · 2026-05-23T17:25:47.102615+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

51 extracted references · 51 canonical work pages

  1. [1]

    Yang, D., K. Xie, K. Ozbay, Z. Zhao, and H. Yang. Copula-based joint modeling of crash count and conflict risk measures with accommodation of mixed count-continuous margins. Analytic Methods in Accident Research, Vol. 31, 2021, p. 100162

    work page 2021

  2. [2]

    E., and R

    Hinton, G. E., and R. R. Salakhutdinov. Reducing the dimensionality of data with neural networks. Science, Vol. 313, No. 5786, 2006, pp. 504-507

    work page 2006

  3. [3]

    Bengio, and G

    LeCun, Y., Y. Bengio, and G. Hinton. Deep learning. Nature, Vol. 521, No. 7553, 2015, pp. 436-444

    work page 2015

  4. [4]

    Ouyang, X

    Liu, L., W. Ouyang, X. Wang, P. Fieguth, J. Chen, X. Liu, and M. Pietikäinen. Deep Learning for Generic Object Detection: A Survey. International Journal of Computer Vision, Vol. 128, No. 2, 2019, pp. 261-318

    work page 2019

  5. [5]

    Sutskever, and G

    Krizhevsky, A., I. Sutskever, and G. E. Hinton. ImageNet classification with deep convolutional neural networks. Communications of the ACM, Vol. 60, No. 6, 2017, pp. 84-90

    work page 2017

  6. [6]

    Cordova-Esparza

    Terven, J., and D. Cordova-Esparza. A Comprehensive Review of YOLO: From YOLOv1 to YOLOv8 and Beyond. arXiv preprint arXiv:2304.00501, 2023

    work page arXiv 2023

  7. [7]

    Ozbay, H

    Xie, K., K. Ozbay, H. Yang, and C. Li. Mining automatically extracted vehicle trajectory data for proactive safety analytics. Transportation research part C: emerging technologies, Vol. 106, 2019, pp. 61-72

    work page 2019

  8. [8]

    Yang, M., G. Han, B. Yan, W. Zhang, J. Qi, H. Lu, and D. Wang. Hybrid-SORT: Weak Cues Matter for Online Multi-Object Tracking. arXiv preprint arXiv:2308.00783, 2023

    work page arXiv 2023

  9. [9]

    Jiang, C

    Zhou, Y., X. Jiang, C. Fu, and H. Liu. Operational factor analysis of the aggressive taxi speeders using random parameters Bayesian LASSO modeling approach. Accid Anal Prev, Vol. 157, 2021, p. 106183

    work page 2021

  10. [10]

    Proactive safety monitoring: A functional approach to detect safety-related anomalies using unmanned aerial vehicle video data

    Yang, D. Proactive safety monitoring: A functional approach to detect safety-related anomalies using unmanned aerial vehicle video data. 2021

    work page 2021

  11. [11]

    Krajewski, T

    Bock, J., R. Krajewski, T. Moers, S. Runde, L. Vater, and L. J. a. e.-p. Eckstein. The inD Dataset: A Drone Dataset of Naturalistic Road User Trajectories at German Intersections.In, 2019. p. arXiv:1911.07602

    work page arXiv 2019

  12. [12]

    Bozcan, I., and E. J. a. e.-p. Kayacan. AU-AIR: A Multi-modal Unmanned Aerial Vehicle Dataset for Low Altitude Traffic Surveillance.In, 2020. p. arXiv:2001.11737

    work page arXiv 2020

  13. [13]

    Abdel-Aty, L

    Zheng, O., M. Abdel-Aty, L. Yue, A. Abdelraouf, Z. Wang, and N. Mahmoud. CitySim: A Drone- Based Vehicle Trajectory Dataset for Safety Oriented Research and Digital Twins. arXiv preprint arXiv:2208.11036, 2022

    work page arXiv 2022

  14. [14]

    Zhu, P., Y. Sun, L. Wen, Y. Feng, and Q. J. a. e.-p. Hu. Drone Based RGBT Vehicle Detection and Counting: A Challenge.In, 2020. p. arXiv:2003.02437

    work page arXiv 2020

  15. [15]

    Geiger, A., P. Lenz, C. Stiller, and R. Urtasun. Vision meets robotics: The KITTI dataset. The International Journal of Robotics Research, Vol. 32, No. 11, 2013, pp. 1231-1237

    work page 2013

  16. [16]

    Fuchs, F. B., A. R. Kosiorek, L. Sun, O. Parker Jones, and I. J. a. e.-p. Posner. End-to-end Recurrent Multi-Object Tracking and Trajectory Prediction with Relational Reasoning.In, 2019. p. arXiv:1907.12887

    work page arXiv 2019

  17. [17]

    Keyvan-Ekbatani, and K

    Guo, H., M. Keyvan-Ekbatani, and K. Xie. Lane change detection and prediction using real-world connected vehicle data. Transportation research part C: emerging technologies, Vol. 142, 2022, p. 103785

    work page 2022

  18. [18]

    Xie, K., H. Yang, X. Dong, H. Yu, and H. Sun. An Automated System for Large-Scale Intersection Marking Data Collection and Condition Assessment.In, 2022

    work page 2022

  19. [19]

    Cetin, H

    Wang, J., M. Cetin, H. Yang, K. Xie, G. Zhai, and S. Ishak. Improving Effectiveness of the Safety Service Patrol Programs: A Discrete Event-Based Simulation Approach. Transportation Research Record, 2024, p. 03611981241248652

    work page 2024

  20. [20]

    Wang, C., F. Guo, R. Yu, L. Wang, and Y. Zhang. The Application of Driver Models in the Safety Assessment of Autonomous Vehicles: A Survey. arXiv preprint arXiv:2303.14779, 2023. Pu, Zhu, Wang, Yang, Xie and Cui 29

    work page arXiv 2023

  21. [21]

    Abdel-Aty, Q

    Gu, X., M. Abdel-Aty, Q. Xiang, Q. Cai, J. J. A. A. Yuan, and Prevention. Utilizing UAV video data for in-depth analysis of drivers’ crash risk at interchange merging areas. Vol. 123, 2019, pp. 159-169

    work page 2019

  22. [22]

    Y., Y.-L

    Chen, A. Y., Y.-L. Chiu, M.-H. Hsieh, P.-W. Lin, and O. J. T. r. p. C. e. t. Angah. Conflict analytics through the vehicle safety space in mixed traffic flows using UAV image sequences. Vol. 119, 2020, p. 102744

    work page 2020

  23. [23]

    Abdel-Aty, O

    Wu, Y., M. Abdel-Aty, O. Zheng, Q. Cai, and S. J. T. r. r. Zhang. Automated safety diagnosis based on unmanned aerial vehicle video and deep learning algorithm. Vol. 2674, No. 8, 2020, pp. 350-359

    work page 2020

  24. [24]

    Ma, Y., H. Meng, S. Chen, J. Zhao, S. Li, and Q. J. J. o. t. e. Xiang, Part A: Systems. Predicting traffic conflicts for expressway diverging areas using vehicle trajectory data. Vol. 146, No. 3, 2020, p. 04020003

    work page 2020

  25. [25]

    Chen, X., Z. Li, Y. Yang, L. Qi, and R. J. I. T. o. I. T. S. Ke. High-resolution vehicle trajectory extraction and denoising from aerial videos. Vol. 22, No. 5, 2020, pp. 3190-3202

    work page 2020

  26. [26]

    Correa, F

    Klautau, A., I. Correa, F. Bastos, I. Nascimento, J. Borges, A. Oliveira, P. Batista, and S. Lins. Integrated simulation of deep learning, computer vision and physical layer of UAV and ground vehicle networks.In Deep Learning and Its Applications for Vehicle Networks, CRC Press. pp. 321-342

    work page

  27. [27]

    Huang, R

    Wang, Y., Z. Huang, R. Laganière, H. Zhang, and L. Ding. A UAV to UAV tracking benchmark. Knowledge-Based Systems, Vol. 261, 2023, p. 110197

    work page 2023

  28. [28]

    Ciaramella, and A

    Di Capua, M., A. Ciaramella, and A. De Prisco. Machine Learning and Computer Vision for the automation of processes in advanced logistics: the Integrated Logistic Platform (ILP) 4.0. Procedia Computer Science, Vol. 217, 2023, pp. 326-338

    work page 2023

  29. [29]

    Liu, and K

    Zhang, N., H. Liu, and K. H. Low. UAV Collision Risk Assessment in Terminal Restricted Area by Heatmap Representation.In AIAA SCITECH 2023 Forum, 2023. p. 0737

    work page 2023

  30. [30]

    Miraki, O

    Gazzea, M., A. Miraki, O. Alisan, M. M. Kuglitsch, I. Pelivan, E. E. Ozguven, and R. Arghandeh. Traffic monitoring system design considering multi-hazard disaster risks. Scientific reports, Vol. 13, No. 1, 2023, p. 4883

    work page 2023

  31. [31]

    Xing, Z., S. Yang, X. Zan, X. Dong, Y. Yao, Z. Liu, and X. Zhang. Flood vulnerability assessment of urban buildings based on integrating high-resolution remote sensing and street view images. Sustainable Cities and Society, Vol. 92, 2023, p. 104467

    work page 2023

  32. [32]

    Petrunin, and D

    Alharbi, A., I. Petrunin, and D. Panagiotakopoulos. Deep Learning Architecture for UAV Traffic- Density Prediction. Drones, Vol. 7, No. 2, 2023, p. 78

    work page 2023

  33. [33]

    Neurohr, T

    Westhofen, L., C. Neurohr, T. Koopmann, M. Butz, B. Schütt, F. Utesch, B. Neurohr, C. Gutenkunst, and E. Böde. Criticality metrics for automated driving: A review and suitability analysis of the state of the art. Archives of Computational Methods in Engineering, Vol. 30, No. 1, 2023, pp. 1-35

    work page 2023

  34. [34]

    Tzutalin

    Tzutalin. Tzutalin. LabelImg. Git code (2015). https://github.com/tzutalin/labelImg . 2015

    work page 2015

  35. [35]

    Ren, S., K. He, R. Girshick, J. J. I. T. o. P. A. Sun, and M. Intelligence. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. Vol. 39, No. 6, 2017, pp. 1137-1149

    work page 2017

  36. [36]

    Anguelov, D

    Liu, W., D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu, and A. C. Berg. SSD: Single Shot MultiBox Detector.In, Springer International Publishing, Cham, 2016. pp. 21-37

    work page 2016

  37. [37]

    Redmon, J., and A. J. a. e.-p. Farhadi. YOLOv3: An Incremental Improvement. 2018

    work page 2018

  38. [38]

    Bochkovskiy, A., C. Y. Wang, and H. Liao. YOLOv4: Optimal Speed and Accuracy of Object Detection. 2020

    work page 2020

  39. [39]

    Gupta, H., O. P. J. M. T. Verma, and Applications. Monitoring and surveillance of urban road traffic using low altitude drone images: a deep learning approach. pp. 1-21

    work page

  40. [40]

    Tian, J., Q. Jin, Y. Wang, J. Yang, S. Zhang, and D. Sun. Performance analysis of deep learning- based object detection algorithms on COCO benchmark: a comparative study. Journal of Engineering and Applied Science, Vol. 71, No. 1, 2024, p. 76

    work page 2024

  41. [41]

    Antwi, R. B., S. Takyi, A. Karaer, E. E. Ozguven, R. Moses, M. A. Dulebenets, and T. Sando. Detecting School Zones on Florida’s Public Roadways Using Aerial Images and Artificial Intelligence (AI2). Transportation Research Record, Vol. 2678, No. 4, 2024, pp. 622-636

    work page 2024

  42. [42]

    Hao, W., and S. Zhili. Improved mosaic: Algorithms for more complex images.In Journal of Physics: Conference Series, No. 1684, IOP Publishing, 2020. p. 012094. Pu, Zhu, Wang, Yang, Xie and Cui 30

    work page 2020

  43. [43]

    Gao, O., C. Niu, W. Liu, T. Li, H. Zhang, and Q. Hu. E-DeepLabV3+: A Landslide Detection Method for Remote Sensing Images.In 2022 IEEE 10th Joint International Information Technology and Artificial Intelligence Conference (ITAIC), No. 10, IEEE, 2022. pp. 573-577

    work page 2022

  44. [44]

    Wang, and L.-P

    Hou, X., Y. Wang, and L.-P. Chau. Vehicle tracking using deep sort with low confidence track filtering.In 2019 16th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS), IEEE, 2019. pp. 1-6

    work page 2019

  45. [45]

    Zhang, Y., P. Sun, Y. Jiang, D. Yu, F. Weng, Z. Yuan, P. Luo, W. Liu, and X. Wang. Bytetrack: Multi-object tracking by associating every detection box.In European Conference on Computer Vision, Springer, 2022. pp. 1-21

    work page 2022

  46. [46]

    Du, Y., Z. Zhao, Y. Song, Y. Zhao, F. Su, T. Gong, and H. Meng. Strongsort: Make deepsort great again. IEEE Transactions on Multimedia, 2023

    work page 2023

  47. [47]

    Khan, Z., S. M. Khan, K. Dey, and M. Chowdhury. Development and evaluation of recurrent neural network-based models for hourly traffic volume and annual average daily traffic prediction. Transportation Research Record, Vol. 2673, No. 7, 2019, pp. 489-503

    work page 2019

  48. [48]

    Tian, Y., and L. Pan. Predicting short-term traffic flow by long short-term memory recurrent neural network.In 2015 IEEE international conference on smart city/SocialCom/SustainCom (SmartCity), IEEE,

    work page 2015

  49. [49]

    Lane, V. M. Obstacle detection and tracking in an urban environment using 3d lidar and a mobileye 560.In, Massachusetts Institute of Technology, 2017

    work page 2017

  50. [50]

    Agarwala, and A

    Vasudevan, V., R. Agarwala, and A. Tiwari. LiDAR-Based Vehicle–Pedestrian Interaction Study on Midblock Crossing Using Trajectory-Based Modified Post-Encroachment Time. Transportation Research Record, 2022, p. 03611981221083295

    work page 2022

  51. [51]

    Fu, C., and H. Liu. Investigating influence factors of traffic violations at signalized intersections using data gathered from traffic enforcement camera. PLoS one, Vol. 15, No. 3, 2020, p. e0229653

    work page 2020