Pith. sign in

REVIEW 4 major objections 5 minor 34 references

PRISA: Proactive Infrastructure LiDAR Framework for Intersection Safety Assessment

T0 review · 4 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read PRISA proposes a modular LiDAR-based intersection safety monitor that runs entirely on edge hardware, auto-curates its own training data, and uses predicted trajectories to flag longitudinal and crossing conflicts before they happen.

desk verdict PRISA is a genuinely modular self-supervised LiDAR safety pipeline, but the real-time feasibility claim rests on a benchmarked alternative stack, not the deployed BlueCity configuration. read the letter →

arxiv 2607.16156 v1 pith:Q23R7EZD submitted 2026-07-17 cs.LG cs.SYeess.SY

Tam Bang , Hussam Abubakr , Emiliano de la Garza Villarreal , Truc Phuong Nguyen , Austin Harris , Toru Hirano , Mina Sartipi , Yunfei Xu
show 1 more author
Hoang H. Nguyen
This is my paper
classification cs.LGcs.SYeess.SY
keywords intersectionsafetymonitoringinfrastructureLiDARsurrogatemeasuresTime-to-CollisionPredictedPost-EncroachmentTimetrajectorypredictionedgedeploymentvulnerableroadusers
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

PRISA's central claim is that proactive intersection safety monitoring no longer requires cloud compute, manual annotation, or clean offline trajectory data. The paper shows a pipeline in which roadside LiDAR tracks road users, automatically cleans accumulated tracks into a training set, trains a trajectory predictor, and feeds forecast trajectories into two conflict metrics—TTC for following vehicles and a predicted post-encroachment time for crossing and vulnerable-road-user interactions. On a public roadside dataset the risk module detects conflicts, and a full deployment on an edge device at a live intersection reports 194 ms end-to-end latency over a 2.4-second prediction horizon. If this is right, intersections can be monitored continuously and cheaply, with warnings issued roughly two seconds before a predicted encounter.

What carries the argument

The load-bearing mechanism is the dual-surrogate-safety assessment: Time-to-Collision (TTC) for longitudinal vehicle-following conflicts, and Predicted Post-Encroachment Time (PPET) for crossing, lane-change, and vulnerable-road-user interactions. PPET generalizes the classical post-encroachment time by propagating each road user's predicted bounding boxes forward in time, detecting overlapping conflict zones in the forecast, and measuring the gap between one user's exit and another's entry. The supporting machinery is a perception-agnostic module: a standardized state vector (position, velocity, heading, box dimensions, class) lets the same risk-assessment code work with any upstream detect

What would settle it

Instrument the actual live-site configuration—the perception backend plus the risk module on the Jetson device—during one hour of peak traffic and measure per-frame end-to-end latency and track completeness. If median latency consistently exceeds the 200 ms frame interval, or if PPET/TTC flags do not match manually reviewed camera footage on a sample of 100 near-miss events, the real-time proactive-safety claim fails.

Watch

Extended reading notes

Core claim

The paper proposes and evaluates PRISA, an end-to-end framework that turns raw roadside LiDAR into conflict alerts. Its distinctive move is to decouple risk assessment from perception behind a standardized state vector, so the same plug-and-play module can consume outputs from any compliant detector and tracker. Site-specific training data are produced automatically: perception outputs are accumulated, filtered, interpolated, and downsampled to form a trajectory database, and a learned trajectory prediction model is trained on it without human labels. In the deployed phase, predicted trajectories feed a dual surrogate-safety evaluation: Time-to-Collision for longitudinal conflicts, and Predi

Load-bearing premise

The deployed perception backend delivers detections fast and accurately enough at the live intersection, since the paper's measured 194 ms latency applies only to a different, fully on-device perception configuration.

Editorial extensions

If this is right

  • Intersection safety monitoring can be automated end-to-end from raw LiDAR to conflict alerts without manual annotation, so new sites can become operational after a short observation period.
  • Because the risk module consumes only a standardized state vector, swapping in a different detector or tracker does not require retraining or re-parameterizing the safety logic.
  • Warnings based on PPET are available before the encounter: with 194 ms processing latency and a 2.4 s horizon, roughly 2.21 s remain for downstream decision-making or alert dissemination.
  • The dual-metric design covers both rear-end and crossing/VRU conflict types; TTC remains unaffected by prediction latency because it is computed from observed states, while PPET inherits the horizon of the trajectory predictor.
  • Deployment on embedded edge hardware avoids cloud offloading, removing variable network delay as a failure mode for deterministic safety-critical operation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The 194 ms figure is measured for a fully on-device perception configuration, not necessarily for the perception backend used at the live site; the real-time claim for the deployed system therefore depends on that backend meeting unmeasured latency and accuracy targets.
  • The reported straight-motion bias in the predictor suggests that some PPET conflict flags at the live site may be false positives; an intent-conditioned predictor or lane-geometry filtering would be a concrete, testable fix.
  • The paper's low Vehicle-VRU conflict counts at the live site (six pairs) are attributed to dedicated pedestrian signal phases; a natural next experiment is deploying the same module at a permissive-turn or mixed-phase intersection, where VRU conflicts are more frequent.
  • Because the data-curation pipeline is automated, conflict detection could run continuously over weeks to build a site-specific near-miss baseline, potentially supporting before-after studies of signal timing or infrastructure changes without new annotation effort.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents PRISA, a modular infrastructure LiDAR framework for real-time intersection safety monitoring. The system consists of a sensing/perception layer and a plug-and-play risk assessment module that (i) self-supervisedly curates site-specific trajectory training data from raw perception outputs, (ii) trains a trajectory prediction model, and (iii) evaluates conflicts using two surrogate safety measures: TTC for longitudinal interactions and PPET for crossing and VRU-involved interactions. The framework is evaluated on the public R-LiViT roadside LiDAR dataset (using ground-truth trajectory annotations as inputs) and deployed at a live signalized intersection in Chattanooga, Tennessee, using an Ouster BlueCity perception backend on an NVIDIA Jetson AGX Thor. The paper claims a 194 ms end-to-end latency and a 2.4 s predictive horizon, arguing practical feasibility for edge-based proactive multi-agent intersection safety monitoring.

Significance. If the claims are fully substantiated, PRISA would be a valuable contribution to the infrastructure-based traffic safety literature: it addresses a real deployment gap by decoupling risk assessment from upstream perception, proposes a self-supervised data-curation pipeline that avoids manual annotation, and demonstrates an edge-hardware deployment. The use of the public R-LiViT benchmark is a welcome step toward reproducibility, and the paper is transparent about several limitations (e.g., fragmented R-LiViT structure, straight-motion bias, elevated VRU–VRU counts). However, the central real-time operational-feasibility claim is currently supported only for a benchmarked on-device perception configuration, not for the actual deployed BlueCity-based system; and the self-supervised curation pipeline is validated only on field data without a clear ground-truth source. These gaps are significant but addressable, making major revision the appropriate recommendation.

major comments (4)
  1. [Section IV.E / Table V] The headline real-time claim (abstract and Section I) rests on the 194 ms end-to-end latency reported in Table V. However, Table V is explicitly for a 'fully on-device configuration' using PointPillars + AB3DMOT. The actual field deployment (Section IV.A.2) uses the Ouster BlueCity API at 10 Hz downsampled to 5 Hz. No latency, throughput, jitter, or execution-location data are reported for the BlueCity configuration. With a 200 ms frame budget at 5 Hz, the 194 ms PointPillars figure leaves essentially no headroom; any additional BlueCity API delay (network, preprocessing, tracking) would push the system past its own cadence. The central feasibility claim is therefore not established for the deployed system.
  2. [Section IV.A.1 / IV.B] The R-LiViT evaluation uses ground-truth trajectory annotations directly as inputs to the risk assessment module, as the paper clearly states. Consequently, the self-supervised data-curation pipeline (Algorithm 1) and its behavior on noisy perception outputs are not validated on a public benchmark. The only validation of the curation pipeline is the field-data ablation in Table VI. This weakens the generalizability claim for the self-supervised training component. The authors should either run the curation pipeline on R-LiViT perception outputs (if available) or explicitly scope the claim to the field deployment and provide stronger evidence for the curation quality.
  3. [Section V / Table VI] Table VI compares trajectory prediction accuracy on 'Ground-Truth Annotations' versus 'Raw Perception + Data Processing' on 4,000 field frames, but the source and construction of the 'Ground-Truth Annotations' are not described. If these annotations were derived from the same BlueCity perception outputs, possibly with manual cleaning, the comparison is not an independent validation of the curation pipeline and the reported 24.1%/26.2% gap could be misleading. Please specify exactly how ground-truth annotations were produced for the field data, and report the number of tracks, evaluation protocol, and any statistical variability.
  4. [Section IV.E] The statement 'TTC is computed directly from observed positions and velocities and is therefore unaffected by prediction latency' is correct only with respect to the trajectory-prediction stage. TTC is still affected by any perception latency in the deployed system: if the BlueCity API delivers states with non-negligible delay, the TTC computation uses stale object positions, and a conflict flagged at time t may already be outdated. The distinction should be clarified: the 194 ms delay affects PPET via prediction latency, but both metrics inherit the perception pipeline's latency.
minor comments (5)
  1. [Section III.B.2 / Algorithm 1] The interpolation step in Algorithm 1 (line 4) is described as 'linear interpolation' for gaps, but the maximum gap length is not specified in Phase I (the filter discards tracks with 'gaps > T_obs', which could be very permissive). Please state the maximum interpolated gap length.
  2. [Table III] Conflict rates exceeding 100% (e.g., Int.3 at 2.5 s TTC) are explained as overlapping multi-pair conflicts. Consider reporting distinct conflict events as well, since rate >100% can confuse readers.
  3. [Equation (6)] TTC uses the absolute difference in scalar velocities. For non-collinear motion, this may overestimate closing speed. It would be helpful to state explicitly that only same-lane following pairs are evaluated, for which this simplification is reasonable.
  4. [Section IV.C] The threshold selection explanation for PPET is somewhat informal. The phrase 'largely saturated at low thresholds' and the choice of 1.5 s as 'more stable' would benefit from a quantitative criterion (e.g., knee detection or false-positive analysis).
  5. [General] The paper uses 'FHW A' in several references (e.g., [7], [8]); please fix the spacing. Also, the access date for the Ouster BlueCity URL is 2025 while the paper is dated 2026; update if appropriate.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; PRISA's derivation chain is independently benchmarked, with one deployment-evidence gap that is a correctness risk, not a circular reduction.

full rationale

PRISA is an empirical system paper rather than a formal derivation, and its load-bearing components are validated against independent anchors. The trajectory prediction models are trained on curated perception data but evaluated on held-out splits (70/30 by chronological order on R-LiViT; separate training/evaluation days at the field site) and compared with external baselines (CV, CA, MID, Trajectron++, FlowChain), so the prediction results are not equal by construction to the training inputs. TTC and PPET are standard definitions applied to predicted trajectories rather than quantities fitted from the data; thresholds are adopted from external FHWA/SSAM literature and a sensitivity sweep, and are not presented as predictions. The self-supervised data-curation paradigm is explicitly stated and is a legitimate training strategy, not a hidden equivalence. The one notable issue, the 194 ms end-to-end latency in Table V being measured for the PointPillars+AB3DMOT configuration while the live deployment used the Ouster BlueCity API with no reported latency, is an evidence/completeness gap about the real-time feasibility claim, not a circularity: no equation or prediction reduces to its own input. Table VI's 'ground-truth annotations' provenance is not stated, but the text provides no indication that those annotations are generated by the same perception pipeline, and per the evidentiary rule a circularity finding requires quoting the specific reduction. There is no load-bearing self-citation chain, no imported uniqueness theorem, and no ansatz smuggled in via citation. The paper is self-contained against a public benchmark and external baselines, so the circularity score is 0.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The framework introduces no new physical entities. Free parameters are dominated by the surfragete safety thresholds chosen via sensitivity analysis on the evaluation data, per-frame horizons, and the speed/ROI constants that shape interaction classification. Assumptions reflect the reliance on prediction model quality, perception accuracy, and the external validity of TTC/PPET thresholds.

free parameters (8)
  • TTC threshold τ_TTC = 1.5 s
    Selected from a sensitivity sweep (0.5-4.0 s) on the same evaluation split; chosen to keep conflict rate below 100% (Section IV.C, Table III).
  • PPET threshold τ_PPET = 1.5 s
    Selected from a sweep (0.5-2.4 s); described as a practical operating point avoiding near-zero PPET artifacts (Section IV.C, Table III).
  • Heading threshold α_long = 30°
    Used to classify following conflicts; adopted from Laureshyn et al. taxonomy (Section III.B.3a).
  • Crossing threshold α_cross = 60°
    Used to classify crossing conflicts; adopted from Laureshyn et al. taxonomy (Section III.B.3a).
  • Speed threshold v_thr = 2.0 m/s
    Used to skip prediction for near-stationary vehicles; chosen without reported justification (Section III.B.2).
  • ROI base radius R_max = 15.0 m
    Base detection radius in the velocity-dependent ROI definition (Section III.B.3a).
  • Observation/prediction horizon T_obs/T_pred = 8 frames / 12 frames
    Follows the FlowChain setup; at 5 Hz this gives 1.6 s observation and 2.4 s prediction (Section IV.B).
  • Frame rate after downsampling = 5 Hz
    BlueCity outputs at 10 Hz, downsampled to 5 Hz to reduce inference frequency; chosen for consistency (Section IV.A.2).
assumptions (5)
  • domain assumption FlowChain (or the chosen trajectory predictor) produces trajectory predictions accurate enough for PPET-based conflict assessment
    The entire PPET pipeline depends on this; the paper does not independently verify predictions against crash outcomes or external near-miss data.
  • domain assumption Ouster BlueCity perception outputs at the field site are accurate and complete enough to serve as training targets and as inputs for conflict detection
    No perception accuracy numbers are reported for the field deployment, yet the self-supervised pipeline and the safety results rely on it.
  • domain assumption Surrogate safety measures (TTC < 1.5 s, PPET < 1.5 s) correspond to genuine safety-critical conflicts
    Adopted from FHWA and the literature; not validated against field conflict annotations in this paper.
  • domain assumption Linear interpolation of gaps up to T_obs in tracks produces physically plausible trajectories
    Algorithm 1 (lines 4-5) interpolates gaps without validation of interpolation fidelity.
  • domain assumption R-LiViT ground-truth annotations are error-free enough for conflict evaluation
    Used as direct inputs to risk assessment in the R-LiViT experiments; no error analysis is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PRISA: Proactive Infrastructure LiDAR Framework for Intersection Safety Assessment." pith.science (2026). https://pith.science/paper/Q23R7EZD

@misc{pith2026260716156,
  author       = {Pith},
  title        = {Pith review of: PRISA: Proactive Infrastructure LiDAR Framework for Intersection Safety Assessment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q23R7EZD}},
  note         = {Machine review of arXiv:2607.16156}
}
read the original abstract

Urban intersections are among the most hazardous locations in road networks, posing significant risks to vehicles and vulnerable road users (VRUs) such as pedestrians and cyclists. The complexity of multi-agent interactions demands continuous, real-time monitoring systems capable of anticipating conflicts before they escalate into crashes. We present PRISA, a modular infrastructure LiDAR framework leveraging privacy-preserving, low-light-robust roadside sensors for long-term traffic observation and real-time risk detection at the edge. The framework comprises two core components: a sensing and perception layer and a plug-and-play risk assessment module. The latter automatically curates site-specific training data from accumulated perception outputs to train a trajectory prediction model without manual annotation. It then deploys the trained model for continuous motion forecasting and dual surrogate safety evaluation, using Time-to-Collision (TTC) for longitudinal conflicts and Predicted Post-Encroachment Time (PPET) for crossing and VRU-involved interactions. PRISA is evaluated on the public R-LiViT dataset and deployed on an NVIDIA Jetson AGX Thor at a live signalized intersection in Chattanooga, Tennessee. PPET-based assessment operates at 194~ms end-to-end latency over a 2.4-second predictive horizon, with TTC-based detection and perception remaining within real-time constraints, demonstrating practical feasibility for proactive multi-agent intersection safety monitoring.

Figures

Figures reproduced from arXiv: 2607.16156 by the authors.

Figure 1
Figure 1. Overview of PRISA, comprising the upstream Sensing [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. RGB camera (top) and LiDAR point cloud (bottom) samples from one representative location within R-LiViT Int.2 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

34 extracted references · 5 linked inside Pith

  1. [1]

    W. H. Organization,Global status report on road safety 2023: country and territory profiles. World Health Organization, 2024

    2023

  2. [2]

    Roadside lidar-based scene understanding toward intelligent traffic per- ception: A comprehensive review,

    J. Zhang, C. Ge, W. Xiao, M. Tang, J. Mills, B. Coifman, and N. Chen, “Roadside lidar-based scene understanding toward intelligent traffic per- ception: A comprehensive review,”ISPRS Journal of Photogrammetry and Remote Sensing, vol. 233, pp. 69–88, 2026

    2026

  3. [3]

    A systematic mapping review of surrogate safety assessment using traffic conflict techniques,

    A. Arun, M. M. Haque, A. Bhaskar, S. Washington, and T. Sayed, “A systematic mapping review of surrogate safety assessment using traffic conflict techniques,”Accident Analysis & Prevention, vol. 153, p. 106016, 2021

    2021

  4. [4]

    A modified post encroachment time model of urban road merging area based on lane-change charac- teristics,

    W. Qi, W. Wang, B. Shen, and J. Wu, “A modified post encroachment time model of urban road merging area based on lane-change charac- teristics,”IEEE Access, vol. 8, pp. 72 835–72 846, 2020

    2020

  5. [5]

    A review of surrogate safety measures uses in historical crash investigations,

    D. Nikolaou, A. Ziakopoulos, and G. Yannis, “A review of surrogate safety measures uses in historical crash investigations,”Sustainability, vol. 15, no. 9, p. 7580, 2023

    2023

  6. [6]

    Surrogate safety assessment in heteroge- neous traffic environment prevailing in developing countries: a system- atic literature review,

    A. Kumar and A. Mudgal, “Surrogate safety assessment in heteroge- neous traffic environment prevailing in developing countries: a system- atic literature review,”International Journal of Injury Control and Safety Promotion, vol. 32, no. 2, pp. 182–200, 2025

    2025

  7. [7]

    Surrogate safety measures from traffic simulation models: Final report,

    D. Gettman and L. Head, “Surrogate safety measures from traffic simulation models: Final report,” Federal Highway Administration, Tech. Rep. FHW A-RD-03-050, 2003

    2003

  8. [8]

    Surrogate safety assess- ment model and validation: Final report,

    D. Gettman, L. Pu, T. Sayed, and S. Shelby, “Surrogate safety assess- ment model and validation: Final report,” Federal Highway Administra- tion, Tech. Rep. FHW A-HRT-08-051, 2008

    2008

Show all 34 references
  1. [9]

    Identifying if VISSIM simulation model and SSAM provide reasonable estimates for field measured traffic conflicts at signalized intersections,

    M. Essa and T. Sayed, “Identifying if VISSIM simulation model and SSAM provide reasonable estimates for field measured traffic conflicts at signalized intersections,”Accident Analysis & Prevention, vol. 45, pp. 684–694, 2013

    2013

  2. [10]

    Conflict-based safety evaluations at unsignalized inter- sections under mixed traffic conditions,

    P. Singhet al., “Conflict-based safety evaluations at unsignalized inter- sections under mixed traffic conditions,”Safety Science, 2024

    2024

  3. [11]

    A review of surrogate safety measures on road infrastructure,

    A. Arunet al., “A review of surrogate safety measures on road infrastructure,”Accident Analysis & Prevention, vol. 191, p. 107222, 2023

    2023

  4. [12]

    A real-time evaluation framework for pedestrian’s potential risk at signalized intersections,

    Y . Liet al., “A real-time evaluation framework for pedestrian’s potential risk at signalized intersections,”arXiv preprint arXiv:2404.15635, 2024

    2024 arXiv

  5. [13]

    Before–after safety analysis of a shared space implementation,

    F. Pascucciet al., “Before–after safety analysis of a shared space implementation,”arXiv preprint arXiv:2307.01058, 2023

    2023 arXiv

  6. [14]

    Stochastic trajectory prediction via motion indeterminacy diffusion,

    T. Gu, G. Chen, J. Li, C. Lin, Y . Rao, J. Zhou, and J. Lu, “Stochastic trajectory prediction via motion indeterminacy diffusion,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 17 113–17 122

    2022

  7. [15]

    Trajectron++: Dynamically-feasible trajectory forecasting with heterogeneous data,

    T. Salzmann, B. Ivanovic, P. Chakravarty, and M. Pavone, “Trajectron++: Dynamically-feasible trajectory forecasting with heterogeneous data,” in European conference on computer vision. Springer, 2020, pp. 683–700

    2020

  8. [16]

    Trajectory prediction for autonomous driving: Progress, limitations, and future directions,

    Y . Huanget al., “Trajectory prediction for autonomous driving: Progress, limitations, and future directions,”arXiv preprint arXiv:2503.03262, 2025

    2025 arXiv

  9. [17]

    Surrogate safety analysis of pedestrian-vehicle conflict at intersections using unmanned aerial vehicle videos,

    P. Chen, W. Zeng, G. Yu, and Y . Wang, “Surrogate safety analysis of pedestrian-vehicle conflict at intersections using unmanned aerial vehicle videos,”Journal of advanced transportation, vol. 2017, no. 1, p. 5202150, 2017

    2017

  10. [18]

    Post encroachment time- based behavioral analysis of vehicle-pedestrian interactions at unsignal- ized midblock crosswalks,

    S. Lamsal, R. Pradhananga, and S. Pandey, “Post encroachment time- based behavioral analysis of vehicle-pedestrian interactions at unsignal- ized midblock crosswalks,”Transportation Research Interdisciplinary Perspectives, vol. 34, p. 101681, 2025

    2025

  11. [19]

    A real-time proactive intersection safety monitoring system based on video data,

    M. Jalayer, N. C. Bouaynaya, D. Patel, and P. Hosseini, “A real-time proactive intersection safety monitoring system based on video data,” 2022

    2022

  12. [20]

    Real-time traffic conflict prediction at signalized intersections using vehicle trajectory data and deep learning,

    G. Zhang, J. Jin, F. Chang, and H. Huang, “Real-time traffic conflict prediction at signalized intersections using vehicle trajectory data and deep learning,”International Journal of Transportation Science and Technology, 2024

    2024

  13. [21]

    Development of a roadside LiDAR-based situational awareness system,

    National Renewable Energy Laboratory, “Development of a roadside LiDAR-based situational awareness system,” U.S. Department of En- ergy, Tech. Rep., 2024

    2024

  14. [22]

    Real-time LiDAR applications for assessing the safety of vulnerable road users at urban intersections,

    SPIE Sensors and Imaging, “Real-time LiDAR applications for assessing the safety of vulnerable road users at urban intersections,” inProceedings of SPIE, vol. 13671, 2025

    2025

  15. [23]

    Enhancing vehicle sensing for traffic safety and mobility performance using roadside LiDAR sensors,

    H. Xuet al., “Enhancing vehicle sensing for traffic safety and mobility performance using roadside LiDAR sensors,” National Transportation Library, Tech. Rep., 2024

    2024

  16. [24]

    G. D. C. Initiative and N. A. of City Transportation Officials,Global street design guide. Island Press, 2016

    2016

  17. [25]

    R-livit: A lidar-visual-thermal dataset enabling vulnerable road user focused roadside perception,

    J. Mirlach, L. Wan, A. Wiedholz, H. E. Keen, and A. Eich, “R-livit: A lidar-visual-thermal dataset enabling vulnerable road user focused roadside perception,”arXiv preprint arXiv:2503.17122, 2025

    2025 arXiv

  18. [26]

    Ouster BlueCity: AI-Driven Lidar for Smarter Cities,

    Ouster Inc., “Ouster BlueCity: AI-Driven Lidar for Smarter Cities,” https://ouster.com/products/software/bluecity, 2025, accessed: Nov. 1, 2025

    2025

  19. [27]

    Fast inference and update of probabilistic density estimation on trajectory prediction,

    T. Maeda and N. Ukita, “Fast inference and update of probabilistic density estimation on trajectory prediction,” inProceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 9795– 9805

    2023

  20. [28]

    What the constant velocity model can teach us about pedestrian motion prediction,

    C. Sch ¨oller, V . Aravantinos, F. Lay, and A. Knoll, “What the constant velocity model can teach us about pedestrian motion prediction,”IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 1696–1703, 2020

    2020

  21. [29]

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

    D. Yang, K. Ozbay, K. Xie, H. Yang, F. Zuo, and D. Sha, “Proactive safety monitoring: A functional approach to detect safety-related anoma- lies using unmanned aerial vehicle video data,”Transportation research part C: emerging technologies, vol. 127, p. 103130, 2021

    2021

  22. [30]

    Evaluation of traffic safety, based on micro-level behavioural data: Theoretical framework and first implementation,

    A. Laureshyn, ˚A. Svensson, and C. Hyd ´en, “Evaluation of traffic safety, based on micro-level behavioural data: Theoretical framework and first implementation,”Accident Analysis & Prevention, vol. 42, no. 6, pp. 1637–1646, 2010

    2010

  23. [31]

    Predicted lane-based time to collision: A novel surrogate safety measure for pedestrian potential risk evaluation at non-signalized intersections,

    T. Lin, Z. Jin, S. Choi, and H. Yeo, “Predicted lane-based time to collision: A novel surrogate safety measure for pedestrian potential risk evaluation at non-signalized intersections,”Transportation Research Interdisciplinary Perspectives, vol. 34, p. 101705, 2025

    2025

  24. [32]

    Pointpillars: Fast encoders for object detection from point clouds,

    A. H. Lang, S. V ora, H. Caesar, L. Zhou, J. Yang, and O. Beijbom, “Pointpillars: Fast encoders for object detection from point clouds,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 12 697–12 705

    2019

  25. [33]

    Lidar AI Solution,

    NVIDIA AI IOT, “Lidar AI Solution,” https://github.com/ NVIDIA-AI-IOT/Lidar AI Solution, 2023, accessed: 2026

    2023

  26. [34]

    Ab3dmot: A baseline for 3d multi-object tracking and new evaluation metrics,

    X. Weng, J. Wang, D. Held, and K. Kitani, “Ab3dmot: A baseline for 3d multi-object tracking and new evaluation metrics,”arXiv preprint arXiv:2008.08063, 2020

    2008 arXiv

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.