Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
184 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
45 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

SIND: A Drone Dataset at Signalized Intersection in China (2209.02297v1)

Published 6 Sep 2022 in cs.CV and cs.GL

Abstract: Intersection is one of the most challenging scenarios for autonomous driving tasks. Due to the complexity and stochasticity, essential applications (e.g., behavior modeling, motion prediction, safety validation, etc.) at intersections rely heavily on data-driven techniques. Thus, there is an intense demand for trajectory datasets of traffic participants (TPs) in intersections. Currently, most intersections in urban areas are equipped with traffic lights. However, there is not yet a large-scale, high-quality, publicly available trajectory dataset for signalized intersections. Therefore, in this paper, a typical two-phase signalized intersection is selected in Tianjin, China. Besides, a pipeline is designed to construct a Signalized INtersection Dataset (SIND), which contains 7 hours of recording including over 13,000 TPs with 7 types. Then, the behaviors of traffic light violations in SIND are recorded. Furthermore, the SIND is also compared with other similar works. The features of the SIND can be summarized as follows: 1) SIND provides more comprehensive information, including traffic light states, motion parameters, High Definition (HD) map, etc. 2) The category of TPs is diverse and characteristic, where the proportion of vulnerable road users (VRUs) is up to 62.6% 3) Multiple traffic light violations of non-motor vehicles are shown. We believe that SIND would be an effective supplement to existing datasets and can promote related research on autonomous driving.The dataset is available online via: https://github.com/SOTIF-AVLab/SinD

Citations (36)

Summary

  • The paper introduces a novel drone dataset that captures over 13,000 trajectories with detailed traffic light and VRU data for autonomous driving research.
  • It employs advanced methods like YOLOv5 detection and Kalman filtering to accurately track and correct trajectories in complex urban scenarios.
  • The dataset uniquely includes traffic light violation data, providing critical insights to predict non-compliant behaviors and enhance vehicle safety systems.

Overview of the Signalized Intersection Drone (SIND) Dataset

The paper "SIND: A Drone Dataset at Signalized Intersection in China" by Yanchao Xu et al., introduces a novel dataset specifically designed for research into autonomous driving technologies at signalized intersections. This dataset, gathered at a signalized intersection in Tianjin, China, leverages high-resolution drone footage to capture real-world traffic scenarios, offering researchers a comprehensive resource for analyzing and modeling autonomous vehicle behavior in complex driving environments.

Dataset Characteristics

The SIND dataset distinguishes itself with several key features that enhance its utility for autonomous driving research:

  • Comprehensive Data: SIND provides extensive details on traffic participants (TPs) including trajectory data, motion parameters, and traffic light states within a high-definition map framework. This holistic data capture allows for a nuanced understanding of dynamic traffic behaviors in response to traffic lights.
  • Diverse Traffic Participant Types: The dataset documents over 13,000 trajectories, encapsulating seven types of TPs, with a significant proportion of Vulnerable Road Users (VRUs) such as pedestrians and cyclists. This focus on VRUs, making up 62.6% of the dataset, addresses a gap seen in other datasets, offering insights into their interaction with motor vehicles in urban environments.
  • Traffic Light Violation Data: Uniquely, SIND includes annotated data on traffic light violations by both motor and non-motor vehicles, which is crucial for developing algorithms that can predict and respond to non-compliant behaviors in real-time.

Methodology

The dataset was created using a systematic and rigorous pipeline that involved:

  • Data Recording and Synchronization: Utilizing drones with 4K cameras, the data was collected over a 15-day period, capturing different weather conditions and times of day. Synchronization with traffic light state recordings was achieved using additional ground-placed cameras.
  • Detection, Tracking, and Correction of Trajectories: Advanced methods like YOLOv5-based detection and intersection-over-union (IOU) matching combined with Kalman filtering were used to identify and track the movement of TPs, while relief displacement corrections were applied to refine these trajectories.
  • Data Processing and Format: The final dataset includes trajectory metadata, detailed motion states, and traffic light states, formatted using lanelet2 map representation, making it practical and applicable for simulation and motion prediction tasks.

Comparative Analysis

The paper provides a comparative analysis of SIND with datasets such as INTERACTION and inD, noting that SIND fills several critical gaps. For instance, while INTERACTION covers multiple locations and scenarios, it lacks comprehensive VRU data and traffic light information. Similarly, inD contains more VRU data, but like INTERACTION, it does not address traffic light states, reducing its applicability for signalized intersection studies.

Implications and Future Work

The SIND dataset is envisaged to have broad implications for research in autonomous vehicle safety and efficacy, particularly in urban environments characterized by mixed traffic scenarios. The detailed trajectory and violation data can enhance predictive models of vehicle behavior, inform safety protocols, and contribute to the development of autonomous navigation systems that better replicate human-like decision-making.

The paper indicates future work will focus on expanding the dataset to additional locations and diverse intersection types, aiming to further support the growing need for varied and comprehensive autonomous driving datasets. Additionally, efforts will be made to generalize and label more complex pedestrian behaviors, increasing the dataset's utility in pedestrian behavior modeling and analysis.

In conclusion, the SIND dataset represents a valuable resource for advancing research in autonomous driving systems, offering thorough and detailed records of intersection dynamics, and setting a new standard for datasets in this field.

Github Logo Streamline Icon: https://streamlinehq.com

GitHub