pith. sign in

arxiv: 1907.07660 · v1 · pith:LBVJABVAnew · submitted 2019-07-17 · 💻 cs.CY · cs.CV

Truck Traffic Monitoring with Satellite Images

Lynn H. Kaack , George H. Chen , M. Granger Morgan This is my paper

Pith reviewed 2026-05-24 19:53 UTC · model grok-4.3

classification 💻 cs.CY cs.CV
keywords satellite imageryobject detectiontruck countingtraffic estimationroad freightemissions monitoringdeveloping countriesmachine learning
0
0 comments X

The pith

An object detection network counts trucks in satellite images to predict average annual daily truck traffic.

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

The paper establishes that satellite images can supply truck traffic data through machine learning where ground monitoring is absent. Road freight drives a rising share of greenhouse gas emissions, yet many low- and middle-income countries lack reliable freight volume records. The work trains an object detection network to count trucks in images, converts those counts into traffic predictions, quantifies uncertainty in the estimates, and examines whether the approach can transfer to data-scarce regions.

Core claim

We show that an object detection network can count trucks in satellite images and that these counts can be used to predict average annual daily truck traffic. We describe a complete model, test the uncertainty of the estimation, and discuss the transfer to developing countries.

What carries the argument

Object detection network that identifies and counts trucks in satellite images to produce traffic volume predictions from the counts.

If this is right

  • Traffic data become available in countries that currently lack ground-based monitoring systems.
  • Road freight volumes and associated emissions can be estimated where surveying is limited.
  • The uncertainty of each traffic prediction can be quantified for decision-making use.
  • The model offers a pathway to extend monitoring into developing countries with limited infrastructure.

Where Pith is reading between the lines

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

  • If the method scales, it could support global tracking of freight emissions without building extensive sensor networks.
  • The approach might be tested on imagery from different satellite sources to check robustness across resolutions.
  • Combining counts with route network data could yield origin-destination estimates for freight flows.

Load-bearing premise

The object detection network accurately identifies trucks in satellite images across regions and conditions, and the counts convert to traffic estimates without major bias from resolution, occlusion, or other image factors.

What would settle it

Direct comparison of the model's predicted traffic volumes against independent ground-based traffic counts collected on the same road segments in a new location would falsify the claim if the predictions show large systematic errors.

Figures

Figures reproduced from arXiv: 1907.07660 by George H. Chen, Lynn H. Kaack, M. Granger Morgan.

Figure 1
Figure 1. Figure 1: A diagram illustrating the overall framework we use for predicting [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: An illustration of the training procedure. We train and select [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Error of truck counts, which also includes false positives, over a [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Cross-validated MAE of different factor regression models to esti [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Out-of-sample traffic variability prediction for three example weeks [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Predicted AADTT from satellite images (box plots) and ground [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Relative count error for each test case grouped by test regions. The [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Predicted AADTT from satellite images (box plots) and ground [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Example images from NY Thruway (a) and BR-116 (b). Green [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Image chips that illustrate what was labeled as a "Truck" indi [PITH_FULL_IMAGE:figures/full_fig_p024_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Schematic illustration of how we computed the NY Thruway [PITH_FULL_IMAGE:figures/full_fig_p026_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: The count data vary between the regions. For example, German [PITH_FULL_IMAGE:figures/full_fig_p027_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Precision-recall curves for validation images on full image (dashed) [PITH_FULL_IMAGE:figures/full_fig_p029_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Count errors as a fraction of the true number of annotated trucks [PITH_FULL_IMAGE:figures/full_fig_p030_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Count errors for the validation set for Brazil. The second model is [PITH_FULL_IMAGE:figures/full_fig_p031_15.png] view at source ↗
read the original abstract

The road freight sector is responsible for a large and growing share of greenhouse gas emissions, but reliable data on the amount of freight that is moved on roads in many parts of the world are scarce. Many low- and middle-income countries have limited ground-based traffic monitoring and freight surveying activities. In this proof of concept, we show that we can use an object detection network to count trucks in satellite images and predict average annual daily truck traffic from those counts. We describe a complete model, test the uncertainty of the estimation, and discuss the transfer to developing countries.

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

1 major / 0 minor

Summary. The manuscript presents a proof-of-concept for using an object detection network to count trucks in satellite images and predict average annual daily truck traffic (AADT) from those counts. It describes a complete model, tests the uncertainty of the estimation, and discusses transfer to developing countries where ground-based freight data are scarce.

Significance. If the mapping from instantaneous counts to AADT proves robust, the approach could supply scalable traffic estimates in regions lacking ground monitoring infrastructure, supporting improved global assessments of road-freight greenhouse-gas emissions.

major comments (1)
  1. [Abstract] Abstract: the central claim requires converting single-moment satellite counts to AADT, yet the description supplies no explicit correction, calibration, or multi-temporal sampling for time-of-day, day-of-week, or seasonal traffic variability; if the uncertainty testing covers only detection error, systematic temporal bias remains unquantified and undermines the prediction step.

Simulated Author's Rebuttal

1 responses · 0 unresolved

Thank you for reviewing our manuscript. We appreciate the feedback highlighting the importance of addressing temporal variability in our AADT estimation. We respond to the major comment as follows.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim requires converting single-moment satellite counts to AADT, yet the description supplies no explicit correction, calibration, or multi-temporal sampling for time-of-day, day-of-week, or seasonal traffic variability; if the uncertainty testing covers only detection error, systematic temporal bias remains unquantified and undermines the prediction step.

    Authors: We agree that the abstract does not explicitly describe the conversion process or address temporal variability. In the manuscript body, we present a proof-of-concept model that uses counts from individual satellite images and applies a scaling factor calibrated against available ground-based AADT data to estimate annual averages. The uncertainty testing primarily quantifies detection errors through repeated annotations and model variations. We acknowledge that systematic biases due to time-of-day or seasonal effects are not fully quantified and represent a limitation of the current approach, particularly given the scarcity of multi-temporal satellite data. We will revise the abstract to clarify the conversion method and note the assumptions involved. Additionally, we will add a dedicated discussion on potential temporal biases and how future work with more frequent imagery could mitigate them. revision: yes

Circularity Check

0 steps flagged

No circularity: standard ML counting pipeline with no self-referential reduction

full rationale

The paper presents a proof-of-concept pipeline that applies an off-the-shelf object detector to satellite imagery, produces truck counts, and then converts those counts into AADT estimates. No equations, fitted parameters, or self-citations are described in the abstract or reader summary that would make any prediction equivalent to its inputs by construction. The conversion step is presented as a modeling choice whose uncertainty is tested, not as a tautology. This is the normal, non-circular case for an applied ML paper.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Ledger is minimal because only the abstract is available; full methods and any fitting details are unknown.

axioms (1)
  • domain assumption Object detection networks can accurately count trucks in satellite images under varied conditions
    This is the core premise enabling the counting step described in the abstract.

pith-pipeline@v0.9.0 · 5613 in / 1059 out tokens · 17552 ms · 2026-05-24T19:53:12.393126+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

12 extracted references · 12 canonical work pages · 3 internal anchors

  1. [1]

    An assessment of remote sensing applications in transportation

    Brent Bowen, Karisa Vlasek, and Cindy Webb. An assessment of remote sensing applications in transportation. In 45th Annual Transportation Research Forum, Evanston, Illinois, March 21-23, 2004, number 208247. Transportation Research Forum,

    work page 2004

  2. [2]

    doi: 10.1109/LGRS.2014.2309695

    ISSN 1545-598X. doi: 10.1109/LGRS.2014.2309695. Zhipeng Deng, Hao Sun, Shilin Zhou, Juanping Zhao, and Huanxin Zou. Toward fast and accurate vehicle detection in aerial images using coupled region-based convolutional neural networks.IEEE Journal of Selected Top- ics in Applied Earth Observations and Remote Sensing,

    work page doi:10.1109/lgrs.2014.2309695 2014

  3. [3]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778,

    work page 2016

  4. [5]

    Speed/accuracy trade-offs for modern convolutional object detectors

    URL http://arxiv.org/abs/1611.10012. Qiling Jiang, Liujuan Cao, Ming Cheng, Cheng Wang, and Jonathan Li. Deep neural networks-based vehicle detection in satellite images. In2015 International Symposium on Bioelectronics and Bioinformatics (ISBB), pages 184–187. IEEE,

    work page internal anchor Pith review Pith/arXiv arXiv

  5. [6]

    xView: Objects in Context in Overhead Imagery

    Darius Lam, Richard Kuzma, Kevin McGee, Samuel Dooley, Michael Laielli, Matthew Klaric, Yaroslav Bulatov, and Brendan McCord. xview: Objects in context in overhead imagery.arXiv preprint arXiv:1802.07856,

    work page internal anchor Pith review Pith/arXiv arXiv

  6. [7]

    URL https://doi.org/10.1080/ 01621459.2017.1307116

    doi: 10.1080/01621459.2017.1307116. URL https://doi.org/10.1080/ 01621459.2017.1307116. Kang Liu and Gellert Mattyus. Fast multiclass vehicle detection on aerial images. IEEE Geoscience and Remote Sensing Letters, 12(9):1938–1942,

    work page doi:10.1080/01621459.2017.1307116 2017

  7. [8]

    People, Penguins and Petri Dishes: Adapting Object Counting Models To New Visual Domains And Object Types Without Forgetting

    URLhttps://arxiv.org/pdf/1711.05586. 20 T Nathan Mundhenk, Goran Konjevod, Wesam A Sakla, and Kofi Boakye. A large contextual dataset for classification, detection and counting of cars with deep learning. In European Conference on Computer Vision, pages 785–800. Springer,

    work page internal anchor Pith review Pith/arXiv arXiv

  8. [9]

    NYS Thruway Origin and Destination Points for All Vehicles - 1 Hour Intervals: 2016, February 2018a

    New York State Thruway Authority. NYS Thruway Origin and Destination Points for All Vehicles - 1 Hour Intervals: 2016, February 2018a. New York State Thruway Authority. NYS Thruway Origin and Destination Points for All Vehicles - 1 Hour Intervals: 2017, February 2018b. New York State Thruway Authority. NYSTA Route System. New York State Thruway Authority,...

    work page 2016

  9. [10]

    World View 3

    21 A Data preparation A.1 Satellite images We use satellite images from DigitalGlobe, Inc., which are taken frequently and by a number of different satellites. We work with 3-channel RGB images of the satellite "World View 3" (VW03), as it has the highest resolution of 31 cm. We found that identifying vehicles in images of the other satellites was difficult....

    work page 2050

  10. [11]

    For the tests, we used the AADTT values from [Caltrans, 2018]

    We ensured ap- proximate compatibility also for other count stations. For the tests, we used the AADTT values from [Caltrans, 2018]. 26 0.0 0.2 0.4 0.6 0.8 1.0 0 100 200 300 Length of count series by station index of dataset days sampled ● NY Thruway California Brazil Germany 0.0 0.2 0.4 0.6 0.8 1.0 0 10000 20000 30000 40000 50000 AADTT by station index o...

    work page 2018

  11. [12]

    complete

    27 A.3.5 Count data summary and sampling To balance between the regions, and ensure sufficiently high AADTT in the training data, we developed our own sampling procedure. We sampled from the stations with the longest count series (most "complete" stations) first, and then sampled from a selection of those with the highest AADTT, to arrive at a dataset of 10 ...

    work page 2016

  12. [13]

    13, we can see that the models performed better when the experiment was constrained to the road

    From Table 2 and Fig. 13, we can see that the models performed better when the experiment was constrained to the road. The full image can contain more difficult examples, for example clustered trucks on parking lots or less typical trucks in junk yards or construction sites. Table 2: Performance on road for optimal prediction probabilityppred; pre- trained ...

    work page 2000