Baike.dev
All toolsAI codingTrendingOpen sourceNewsSubmit
Log in
< Back to tools
S

SinD

> 编程语言
Open source

SinD: a drone-based trajectory dataset for heterogeneous traffic at signalized intersections in China

197 stars0 likes0 views
WebsiteGitHub

About

SinD: a drone-based trajectory dataset for heterogeneous traffic at signalized intersections in China

SinD Dataset (Signalized Intersection Drone Dataset)

Welcome to the official repository of the SinD Dataset, a drone-based trajectory dataset for heterogeneous traffic at signalized intersections in China.

SinD v2.0 is now available. The new release features cross-domain intersection diversity, high-interaction traffic, semantic scenario annotations, a compact policy-testing toolchain, and a 3DGS visual simulation extension.

Paper · YouTube Demo · Bilibili Demo


Table of Contents

  • SinD v2.0 (Latest)
    • What's New in v2.0?
  • Data Subset
    • SIND Tianjin
    • SIND Chongqing
    • SIND Changchun
    • SIND Xi'an
  • SinD 2.0 Dataset Feature Analysis
  • Semantic Scenario Examples
    • High-risk MprTTC interactions
    • Visual shielding
    • Narrow feasible area
    • Rule-violation cases
    • Scenario Testing Toolchain
    • 3DGS Visual Simulation Extension
    • SinD v2.0 Resources
  • SinD v1.0
  • Quick Installation
  • Dataset Access
  • Citation
  • Acknowledgements

SinD v2.0 (Latest)

What's New in v2.0?

  • Cross-domain diversity: trajectories from six signalized intersections across four cities: Changchun, Tianjin, Xi'an, and Chongqing.
  • High-interaction traffic: heterogeneous participants, dense multi-agent negotiation, and location-dependent kinematic regimes.
  • Semantic scenario annotations: SOTIF-oriented labels for high-risk interaction, visual shielding, narrow feasible areas, and rule-related events.
  • Testing toolchain: reproducible scene replay and policy testing with ground truth, RiskIDM, and Diffusion baselines, with the integration of trajdata.
  • 3DGS extension (will be released soon): reconstructed intersections and ego-view simulation for camera-based and vision-based autonomous-driving research.

Data Subset

SIND_Tianjin

SIND_Tianjin contains 7 hours of recording including over 13,000 traffic participants with 7 types, HD maps and traffic light information are used to count traffic light violations by vehicles in them. Clearly, SIND_Tianjin has a high proportion of vulnerable road users and frequent non-motor vehicle violations.

SIND_Chongqing

Sind_Chongqing was collected at an intersection in Chongqing, where the traffic density was low and the freedom of traffic participants was high; Compared to the situation where pedestrians and vehicles share traffic lights in Sind_Tianjin, Sind_Chongqing have independent vehicle traffic lights and pedestrian traffic lights. However, for vehicles, the conflict between turning left and going straight is still normal.

SIND_Changchun

SinD_Changchun is an intersection located on a traffic artery with a high traffic density in Changchun. In this dataset, dense unprotected left turn conflicts can be observed, and even conflicting traffic congestion occurs when a green wave of traffic cannot completely pass through the intersection.

SIND_Xi'an

SinD_Xi'an was collected at an intersection with moderate traffic density in Xi'an City, mainly consisting of vehicles; Similar to SinD-Tianjin, it has a shared traffic signal for pedestrians and vehicles, and there are conflicts between left turns and straight traffic.

SinD 2.0 Dataset Feature Analysis

SinD 2.0 provides precomputed cross-intersection analysis examples for scenario understanding and benchmark design. The interaction-degree summary measures whether each interaction component contains two, three, or four-plus traffic participants, exposing how often scenes require multi-agent reasoning rather than pairwise conflict handling.

The kinematic envelope compares the 95% speed-acceleration operating regions across the six released SinD 2.0 intersections. It highlights location-dependent driving styles and corner-case motion regimes that are useful for domain-shift analysis, policy stress testing, and scenario mining.

The conflict-mode chord diagram summarizes how maneuver pairs contribute to interaction episodes across city groups. It provides a compact view of site-specific conflict structure, such as left-turn versus straight-through interactions and vulnerable-road-user conflicts.

Semantic Scenario Examples

SinD 2.0 also includes semantic scenario labels for structured high-value cases. The public schema supports a shared time window and ego agent, plus type-specific fields such as challenger, shielding, shielded, risk score, feasible-area, and violation metadata.

The semantic labeling framework uses a shared scenario schema for common metadata and extends it with event-specific fields for non-compliance, high-risk interaction, visual shielding, and narrow feasible-area triggers.

High-risk MprTTC interactions

These cases mark short-horizon interaction risks between an ego participant and a primary conflicting participant, while preserving surrounding traffic context for policy evaluation.

Visual shielding

Visual-shielding cases describe occlusion relationships among ego, occluding participant, and hidden target, capturing situations where conflict risk is revealed only after partial emergence.

Narrow feasible area

Narrow feasible-area cases identify scenes where surrounding participants constrain the ego's reachable space, exposing dense traffic negotiation and blocked front-grid regions.

Rule-violation cases

Rule-violation examples cover red-light entry, yellow-light entry, lane-direction violation, solid-line lane change, wrong-way candidates, and vulnerable-road-user encroachment.

Representative labels are provided in datasets/SinD_dataset/Semantic_labels/scenarios_sample.json, and the browser preview in semantic_preview/index.html can load either the sample file or a full local scenarios.json. Full labels can be accessed by applying for full dataset.

Scenario Testing Toolchain

The public toolchain turns an ordinary scene or semantic clip into a reproducible test project with trajectory replay (Open-Loop Test), policy control (Closed-loop Test / Interactive Test), risk metrics, logs, and interactive result visualization.

Public policy Role
ground_truth Naturalistic trajectory replay
risk_idm Risk-extended IDM baseline
diffuser Diffusion baseline; model weights are provided after an approved application

See the testing toolchain guide for runnable templates and environment requirements.

3DGS Visual Simulation Extension

The 3DGS extension reconstructs signalized intersections, inserts auxiliary traffic assets, and renders controlled ego-view sequences for vision-based autonomous-driving evaluation.

The extension is going to be prepared as an auxiliary release soon.

SinD v2.0 Resources

Resource Link
Data layouts, conversion, and preprocessing DATASETS.md
Semantic label schema and video preview semantic_preview/README.md
RiskIDM/Diffusion testing toolchain Simulation_test_toolchain/README.md
Python dataset interface src/trajdata/

️ SinD v1.0

SinD v1.0 is the original dataset presented at ITSC 2022. The repository keeps its public sample records, CSV format specification, maps, and desktop visualization tool for backward compatibility.

Legacy resource Link
Public sample records Data/
Original CSV format Format.md
Original visualization tool SIND-Vis-tool/
ITSC 2022 paper arXiv:2209.02297

Quick Installation

conda create -n sind2 python=3.10
conda activate sind2
pip install -r requirements.txt
pip install -e .

Lanelet2 is optional but recommended for full map functionality. Detailed data preparation instructions are provided in DATASETS.md.


Dataset Access

To request the full dataset, contact [email protected], [email protected], [email protected], or [email protected] using an educational email address.

Use the subject:

[Apply for SinD] name_country(region)_organization

Please include your laboratory or department, research interests, and intended use of the dataset. If you need the Diffuser model weights, explicitly state this in your application.


Citation

If you find the dataset or toolchain useful, please cite the corresponding paper.

SinD v2.0 (arXiv:2607.16943)

@article{sindv2_2026,
  title={SinD 2.0: A Multi-City UAV Dataset with Semantic Risk Annotations for SOTIF-Oriented Safety Validation at Signalized Intersections},
  author={Li, Yunwei and Fu, Shengjie and Chen, Chunrong and Zhao, Chengxiang and Fan, Yuchen and Zhu, Mingyu and Xu, Yanchao and Zhang, Yuxin and Yang, Lan and Li, Chuzhao and Ji, Jie and He, Yi and Sarkar, Abhijit and Sonth, Akash and Wang, Hong and Li, Jun},
  journal={arXiv preprint arXiv:2607.16943},
  year={2026}
}

SinD v1.0 (ITSC 2022)

@INPROCEEDINGS{9921959,
  author={Xu, Yanchao and Shao, Wenbo and Li, Jun and Yang, Kai and Wang, Weida and Huang, Hua and Lv, Chen and Wang, Hong},
  booktitle={2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC)},
  title={SIND: A Drone Dataset at Signalized Intersection in China},
  year={2022},
  pages={2471-2478},
  doi={10.1109/ITSC55140.2022.9921959}
}

If you use the trajdata interface, please also cite the trajdata paper where appropriate.


Acknowledgements

The original visualization code builds on the public tooling of the inD dataset and the INTERACTION dataset.

  • School of Vehicle and Mobility, Tsinghua University
  • Tsinghua Intelligent Vehicle Design and Safety Research Institute
  • Safety Of The Intended Functionality (SOTIF) Research Team

Issues· 0 open

View all issuesOpen on GitHub

No open issues yet, or sync has not completed.

> Tags

Python

No comments yet. Be the first to share.

> Details

PublishedAug 1, 2026
UpdatedSep 18, 2026
Category编程语言
PricingOpen source

> Related tools

T
TypeScript
JavaScript 的超集,为前端与全栈提供静态类型
P
Python
通用编程语言,广泛用于 Web、数据与 AI
G
Go
Google 推出的简洁高效系统语言