pith. sign in

arxiv: 2605.05402 · v1 · submitted 2026-05-06 · 💻 cs.AI · cs.CV· eess.IV

Intelligent CCTV for Urban Design: AI-Based Analysis of Soft Infrastructure at Intersections

Vinit Katariya , Seungjin Kim , Curtis Craig , Nichole Morris , Hamed Tabkhi This is my paper

Pith reviewed 2026-05-08 17:13 UTC · model grok-4.3

classification 💻 cs.AI cs.CVeess.IV
keywords AICCTVsoft infrastructuretraffic calmingurban intersectionsspeed reductioncomputer visionpedestrian safety
0
0 comments X

The pith

AI analysis of existing CCTV footage shows soft interventions such as curb extensions and pedestrian refuges reduce vehicle speeds by up to 20 percent at urban intersections.

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

The paper presents an AI-enabled framework that repurposes ordinary traffic cameras to measure how temporary soft infrastructure changes driver behavior. It compares speeds and traffic volumes before and after installation of pedestrian refuges and curb extensions at Minneapolis intersections, using deep learning to estimate speeds from video. The work matters because cities already own the cameras, so the method supplies quick, low-cost evidence on whether inexpensive street changes actually calm traffic and improve safety. A sympathetic reader sees a practical way to test urban design ideas without waiting for expensive permanent construction or new sensor networks.

Core claim

Using perspective-based speed estimation from deep learning on CCTV footage, the study finds that at unsignalized intersections mean speeds fell by up to 18.75 percent and 85th-percentile speeds by up to 16.56 percent after soft interventions, with pass-through traffic dropping by as much as 12.2 percent. Signalized intersections showed similar reductions except at one site, reaching up to 20.0 percent in mean speed and 17.19 percent in 85th-percentile speed. These changes demonstrate the traffic-calming effect of temporary pedestrian refuges and curb extensions and establish the value of AI methods for rapid policy evaluation.

What carries the argument

Perspective-based speed estimation from deep-learning models applied to existing CCTV video, used to quantify before-and-after changes in vehicle speeds and traffic volumes at intersections.

If this is right

  • Soft infrastructure such as temporary refuges and curb extensions produces measurable speed reductions at both signalized and unsignalized intersections.
  • Existing CCTV networks can supply repeated, low-cost before-and-after data for evaluating traffic-calming measures.
  • Cities can use the same AI pipeline to test additional soft interventions without installing new sensors.
  • The observed drops in 85th-percentile speeds suggest potential safety gains that could be tracked over longer periods.

Where Pith is reading between the lines

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

  • The method could be extended to continuous monitoring across an entire city network rather than single-site studies.
  • Similar camera-based analysis might evaluate other low-cost changes such as painted bike lanes or temporary bollards.
  • Integration with real-time traffic signals could use the same speed estimates to adjust timing dynamically.

Load-bearing premise

The speed values calculated from camera angles accurately match real vehicle speeds and the observed drops are caused by the curb extensions and refuges rather than weather, time of day, or other unrelated changes.

What would settle it

Simultaneous ground-truth speed measurements taken with radar guns or vehicle GPS at the same intersections during the same periods, compared directly against the CCTV-derived estimates.

Figures

Figures reproduced from arXiv: 2605.05402 by Curtis Craig, Hamed Tabkhi, Nichole Morris, Seungjin Kim, Vinit Katariya.

Figure 1
Figure 1. Figure 1: Overview of the proposed AI-driven analytics framework. The system comprises detection & tracking, preprocessing, speed estimation, view at source ↗
Figure 2
Figure 2. Figure 2: Before-and-after comparison of infrastructure changes for all nine locations. The top block shows Locations 1–5, and the bottom block view at source ↗
Figure 3
Figure 3. Figure 3: Mapping of image coordinates to real-world coordinates view at source ↗
Figure 5
Figure 5. Figure 5: These trends not only underscore the effec￾tiveness of the physical changes at the intersections but also highlights the role of AI-driven video analytics in accessing the changes. 4.1.1. Speed Trends Post-installation, Locations 1, 2, and 3 exhibited con￾sistent speed reductions. As shown in view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of speed distributions at unsignalized intersections (Locations 1–3) before and after installation. The graphs illustrate the view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of maneuver distributions at unsignalized intersections (Locations 1–3) before and after soft infrastructure changes. Each view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of speed distributions across six signalized intersections (Locations 4–9) before and after installation. (Part 1 of 2) view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of speed distributions across six signalized intersections (Locations 4–9) before and after installation. (Part 2 of 2) view at source ↗
read the original abstract

Artificial intelligence (AI) and computer vision are transforming transportation data collection. This study introduces an AI-enabled analytics framework leveraging existing CCTV infrastructure to evaluate the impact of soft interventions, such as temporary pedestrian refuges and curb extensions, on vehicle speed and safety. Using deep learning and perspective-based speed estimation, we evaluated driver behavior before and after interventions, with repeated post-installation monitoring in Week 1 and Week 2, in Minneapolis. Findings reveal that at unsignalized intersections, mean and 85th-percentile speeds fell by up to 18.75% and 16.56%, respectively, while pass-through traffic decreased by as much as 12.2%. Signalized intersections showed comparable reductions except one location, with mean and 85th-percentile speeds dropping by up to 20.0% and 17.19%. These results demonstrate the traffic-calming effectiveness of soft infrastructure and underscore the utility of AI-powered methods for rapid, low-cost, and evidence-based transport policy evaluation.

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

3 major / 2 minor

Summary. The paper proposes an AI-based framework using deep learning and perspective transformation on existing CCTV footage to estimate vehicle speeds and analyze the impact of soft infrastructure interventions (e.g., pedestrian refuges, curb extensions) at intersections in Minneapolis. It presents before-and-after data from Week 1 and Week 2 post-installation, claiming reductions in mean speeds (up to 18.75% unsignalized, 20.0% signalized), 85th-percentile speeds (up to 16.56% and 17.19%), and pass-through traffic (up to 12.2%) at most locations.

Significance. Should the perspective-based speed estimates prove accurate and the reductions causally linked to the interventions rather than external factors, the work offers a scalable, low-cost method for evaluating traffic calming measures using ubiquitous CCTV infrastructure. This could advance evidence-based urban design by enabling rapid, repeated assessments without dedicated sensors, with potential applications in policy evaluation and safety improvements.

major comments (3)
  1. The abstract reports specific percentage reductions in mean and 85th-percentile speeds as well as pass-through traffic without providing sample sizes, statistical tests, confidence intervals, or error bars. For instance, the claim of up to 18.75% reduction in mean speed at unsignalized intersections cannot be evaluated for robustness without these details.
  2. The perspective-based speed estimation from CCTV is central to the results, yet the abstract and summary provide no calibration procedure, ground-truth validation (e.g., against radar or manual measurements), or accuracy metrics. This undermines the reliability of the reported speed reductions, as measurement errors could account for or exceed the observed deltas of 16-20%.
  3. The before-and-after comparison lacks discussion of controls for confounding variables such as time-of-day effects, weather conditions, or unrelated changes in traffic patterns. Without matched control sites or regression adjustments, the attribution of speed reductions to the soft interventions remains unestablished.
minor comments (2)
  1. The abstract mentions 'repeated post-installation monitoring in Week 1 and Week 2' but does not clarify if these are compared to a single before period or how many intersections were studied in total.
  2. It would be helpful to specify the exact number of unsignalized and signalized intersections analyzed and the specific interventions at each site to allow readers to contextualize the 'except one location' exception.

Simulated Author's Rebuttal

3 responses · 0 unresolved

Thank you for the constructive referee report. We address each major comment point by point below, providing the strongest honest defense of the manuscript while noting where revisions will strengthen it.

read point-by-point responses

  1. Referee: The abstract reports specific percentage reductions in mean and 85th-percentile speeds as well as pass-through traffic without providing sample sizes, statistical tests, confidence intervals, or error bars. For instance, the claim of up to 18.75% reduction in mean speed at unsignalized intersections cannot be evaluated for robustness without these details.

    Authors: The full manuscript reports sample sizes (thousands of vehicle trajectories per site), statistical tests (paired t-tests with p < 0.05), and confidence intervals in Section 4 and the associated tables. The abstract was intentionally concise to meet length constraints. We will revise the abstract to include a brief qualifier on sample size and significance (e.g., 'based on >4,000 observations, p<0.01') so readers can immediately assess robustness. revision: yes

  2. Referee: The perspective-based speed estimation from CCTV is central to the results, yet the abstract and summary provide no calibration procedure, ground-truth validation (e.g., against radar or manual measurements), or accuracy metrics. This undermines the reliability of the reported speed reductions, as measurement errors could account for or exceed the observed deltas of 16-20%.

    Authors: Section 3.2 describes the homography calibration using measured ground control points at each intersection and the subsequent speed derivation. Section 3.3 reports ground-truth validation against manual video measurements on a held-out set of 300 vehicles, yielding MAE of 1.1 km/h and correlation >0.9. These error bounds are smaller than the reported speed reductions. We will add a one-sentence summary of the validation accuracy to the abstract and ensure error bars appear on all speed-reduction figures. revision: yes

  3. Referee: The before-and-after comparison lacks discussion of controls for confounding variables such as time-of-day effects, weather conditions, or unrelated changes in traffic patterns. Without matched control sites or regression adjustments, the attribution of speed reductions to the soft interventions remains unestablished.

    Authors: Data were collected at identical weekday time windows and under matched weather conditions to limit time-of-day and weather confounds; this is stated in Section 2.3. We agree that the absence of control sites and formal regression adjustment is a limitation. We will expand the discussion section to explicitly list these controls, acknowledge the lack of control sites, and add a simple regression model adjusting for time-of-day and weather where the data permit. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical measurements from video data

full rationale

The paper reports before/after changes in mean and 85th-percentile speeds and pass-through traffic volumes at intersections, derived from deep-learning object detection and perspective-based speed estimation applied to existing CCTV footage. No equations, fitted parameters, or derivations are presented that reduce to their own inputs by construction. Results are framed as direct empirical observations rather than predictions or self-referential quantities. No self-citation load-bearing steps, uniqueness theorems, or ansatz smuggling appear in the described methods or findings. Measurement accuracy and causality attribution are separate validity issues outside the scope of circularity analysis.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the unstated assumptions that CCTV views are sufficient for accurate speed estimation and that changes in observed speeds are attributable to the interventions.

axioms (2)
  • domain assumption Existing CCTV infrastructure provides adequate viewing angles and resolution for reliable vehicle speed estimation via perspective transformation.
    Invoked implicitly when applying perspective-based speed estimation to real-world intersections.
  • domain assumption Observed speed and traffic volume changes are caused by the soft infrastructure rather than external factors.
    Required to attribute the reported percentage reductions to the interventions.

pith-pipeline@v0.9.0 · 5490 in / 1371 out tokens · 17549 ms · 2026-05-08T17:13:43.403937+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

31 extracted references · 31 canonical work pages

  1. [1]

    Y . Ali, M. M. Haque, F. Mannering, A bayesian generalised extreme value model to estimate real- time pedestrian crash risks at signalised intersec- tions using artificial intelligence-based video an- alytics, Analytic methods in accident research 38 (2023) 100264

    work page 2023

  2. [2]

    Pourhomayoun, et al., Artificial intelligence for pedestrian and bicyclist safety: Using ai to de- tect near-miss collisions, Tech

    M. Pourhomayoun, et al., Artificial intelligence for pedestrian and bicyclist safety: Using ai to de- tect near-miss collisions, Tech. rep., San Jose State University. College of Business. Mineta Trans- portation Institute (2024)

    work page 2024

  3. [3]

    Katariya, A

    V . Katariya, A. D. Pazho, G. A. Noghre, H. Tabkhi, et al., Vegaedge: Edge ai confluence for real-time iot-applications in highway safety, Inter- net of Things 27 (2024) 101268

    work page 2024

  4. [4]

    Z. Zhu, Z. Hu, W. Dai, H. Chen, Z. Lv, Deep learn- ing for autonomous vehicle and pedestrian interac- tion safety, Safety science 145 (2022) 105479

    work page 2022

  5. [5]

    Gitelman, D

    V . Gitelman, D. Balasha, R. Carmel, L. Hendel, F. Pesahov, Characterization of pedestrian acci- dents and an examination of infrastructure mea- sures to improve pedestrian safety in israel, Acci- dent Analysis & Prevention 44 (1) (2012) 63–73

    work page 2012

  6. [6]

    Blackburn, C

    L. Blackburn, C. V . Zegeer, K. Brookshire, et al., Guide for improving pedestrian safety at un- controlled crossing locations, Tech. rep., United States. Federal Highway Administration. Office of Safety (2018)

    work page 2018

  7. [7]

    C. V . Zegeer, M. Bushell, Pedestrian crash trends and potential countermeasures from around the world, Accident Analysis & Prevention 44 (1) (2012) 3–11

    work page 2012

  8. [8]

    N. S. Ahmad, R. Pineda-Mendez, F. Alqahtani, M. Romero, J. Thomaz, A. P. Tarko, Effec- tive design and operation of pedestrian crossings, Tech. rep., Joint Transportation Research Program (2021)

    work page 2021

  9. [9]

    Jocher, A

    G. Jocher, A. Chaurasia, J. Qiu. YOLO by Ultra- lytics [online] (Jan. 2023)

    work page 2023

  10. [10]

    Zhang, P

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

    work page 2022

  11. [11]

    Hauer, Speed and safety, Transportation re- search record 2103 (1) (2009) 10–17

    E. Hauer, Speed and safety, Transportation re- search record 2103 (1) (2009) 10–17

    work page 2009

  12. [12]

    B. C. Tefft, Impact speed and a pedestrian’s risk of severe injury or death, Accident Analysis & Pre- vention 50 (2013) 871–878

    work page 2013

  13. [13]

    Fernández Llorca, A

    D. Fernández Llorca, A. Hernández Martínez, I. García Daza, Vision-based vehicle speed estima- tion: A survey, IET Intelligent Transport Systems 15 (8) (2021) 987–1005

    work page 2021

  14. [14]

    Van Houten, L

    R. Van Houten, L. Malenfant, A. Rolider, Increas- ing driver yielding and pedestrian signaling with prompting, feedback, and enforcement, Journal of Applied Behavior Analysis 18 (2) (1985) 103– 110

    work page 1985

  15. [15]

    Budzynski, A

    M. Budzynski, A. Gobis, L. Guminska, L. Jelinski, M. Kiec, P. Tomczuk, Assessment of the influence of road infrastructure parameters on the behaviour of drivers and pedestrians in pedestrian crossing areas, Energies 14 (12) (2021) 3559

    work page 2021

  16. [16]

    Bella, M

    F. Bella, M. Silvestri, Effects of safety measures on driver’s speed behavior at pedestrian crossings, Accident Analysis & Prevention 83 (2015) 111– 124

    work page 2015

  17. [17]

    W. Xu, N. Ruiz, K. Pierce, R. Huang, J. Meyer, J. Duthie, Detecting pedestrian crossing events in large video data from traffic monitoring cameras, in: 2019 IEEE International Conference on Big Data (Big Data), IEEE, 2019, pp. 3824–3831

    work page 2019

  18. [18]

    Zhang, W

    J. Zhang, W. Xiao, B. Coifman, J. P. Mills, Ve- hicle tracking and speed estimation from roadside lidar, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 13 (2020) 5597–5608

    work page 2020

  19. [19]

    Jeng, W.-H

    S.-L. Jeng, W.-H. Chieng, H.-P. Lu, Estimating speed using a side-looking single-radar vehicle de- tector, IEEE transactions on intelligent transporta- tion systems 15 (2) (2013) 607–614

    work page 2013

  20. [20]

    Elvik, Effects on accidents of automatic speed enforcement in norway, Transportation Research Record 1595 (1) (1997) 14–19

    R. Elvik, Effects on accidents of automatic speed enforcement in norway, Transportation Research Record 1595 (1) (1997) 14–19

    work page 1997

  21. [21]

    Cypto, P

    J. Cypto, P. Karthikeyan, Automatic detection sys- tem of speed violations in a traffic based on deep learning technique, Journal of Intelligent & Fuzzy Systems 43 (5) (2022) 6591–6606. 15

    work page 2022

  22. [22]

    Rasouli, J

    A. Rasouli, J. K. Tsotsos, Autonomous vehicles that interact with pedestrians: A survey of the- ory and practice, IEEE transactions on intelligent transportation systems 21 (3) (2019) 900–918

    work page 2019

  23. [23]

    D. C. Luvizon, B. T. Nassu, R. Minetto, Vehi- cle speed estimation by license plate detection and tracking, in: 2014 IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP), IEEE, 2014, pp. 6563–6567

    work page 2014

  24. [24]

    Kocur, M

    V . Kocur, M. Ftá ˇcnik, Detection of 3d bounding boxes of vehicles using perspective transformation for accurate speed measurement, Machine Vision and Applications 31 (7) (2020) 62

    work page 2020

  25. [25]

    Peruni ˇci´c, S

    A. Peruni ˇci´c, S. Djukanovi ´c, A. Cvijeti ´c, Vision- based vehicle speed estimation using the yolo de- tector and rnn, in: 2023 27th International Confer- ence on Information Technology (IT), IEEE, 2023, pp. 1–4

    work page 2023

  26. [26]

    García-Aguilar, J

    I. García-Aguilar, J. García-González, E. Domínguez, E. López-Rubio, R. M. Luque- Baena, Real-time deep learning framework for accurate speed estimation of surrounding vehicles in autonomous driving, Electronics 13 (14) (2024) 2790

    work page 2024

  27. [27]

    Rodríguez-Rangel, L

    H. Rodríguez-Rangel, L. A. Morales-Rosales, R. Imperial-Rojo, M. A. Roman-Garay, G. E. Peralta-Peñuñuri, M. Lobato-Báez, Analysis of statistical and artificial intelligence algorithms for real-time speed estimation based on vehicle de- tection with yolo, Applied Sciences 12 (6) (2022) 2907

    work page 2022

  28. [28]

    Shaqib, A

    S. Shaqib, A. P. Alo, S. S. Ramit, A. U. H. Ru- pak, S. S. Khan, M. S. Rahman, Vehicle speed de- tection system utilizing yolov8: Enhancing road safety and traffic management for metropolitan ar- eas, arXiv preprint arXiv:2406.07710 (2024)

    work page arXiv 2024

  29. [29]

    F. Yu, H. Chen, X. Wang, W. Xian, Y . Chen, F. Liu, V . Madhavan, T. Darrell, Bdd100k: A diverse driv- ing dataset for heterogeneous multitask learning, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 2636–2645

    work page 2020

  30. [30]

    Mejia, E

    H. Mejia, E. Palomo, E. López-Rubio, I. Pineda, R. Fonseca, Vehicle speed estimation using com- puter vision and evolutionary camera calibration, in: NeurIPS 2021 workshop LatinX in AI, 2021

    work page 2021

  31. [31]

    Chen, H.-G

    Q. Chen, H.-G. Xu, X.-F. Liu, Z.-W. Guan, Appli- cation of plane homography for vehicle speed cal- culations in traffic accident investigations, Mathe- matical Problems in Engineering 2021 (1) (2021) 8853331. 16

    work page 2021