V-TIDB (VARIOUS-PERSPECTIVE TRAFFIC INCIDENT DATABASE)

paper: Traffic Incident Database with Multiple Labels Including Various Perspective Environmental Information (IROS2023)
Author: Shota Nishiyama, Takuma Saito, Ryo Nakamura, Hirokatsu Kataoka, Kensho Hara
Affiliation, Organization:
Aichi Institute of Technology Graduate School of Business Administration and Computer Science (AIT),
Tokyo Denki University,
National Institute of Advanced Industrial Science and Technology (AIST),
Fukuoka University,
Keio University

Overview

V-TIDB is a dataset with 10 different labels with various perspectives, and the effectiveness of the various perspective environmental information was demonstrated in an experiment using 3D ResNet.

Abstract

Abstract— Traffic accident recognition is essential in develop- ing automated driving and Advanced Driving Assistant System technologies. A large dataset of annotated traffic accidents is necessary to improve the accuracy of traffic accident recogni- tion using deep learning models. Conventional traffic accident datasets provide annotations on the presence or absence of traffic accidents and other teacher labels, improving traffic ac- cident recognition performance. However, the labels annotated in conventional datasets need to be more comprehensive to de- scribe traffic accidents in detail. Therefore, we propose V-TIDB, a large-scale traffic accident recognition dataset annotated with various environmental information as multi-labels. Our proposed dataset aims to improve the performance of traffic accident recognition by annotating ten types of environmental information as teacher labels in addition to the presence or absence of traffic accidents. V-TIDB is constructed by collecting many videos from the Internet and annotating them with appropriate environmental information. In our experiments, we compare the performance of traffic accident recognition when only labels related to the presence or absence of traffic accidents are trained and when environmental information is added as a multi-label. In the second experiment, we compare the performance of the training with only “contact level,” which represents the severity of the traffic accident, and the performance with environmental information added as a multi- label. The results showed that 6 out of 10 environmental information labels improved the performance of recognizing the presence or absence of traffic accidents. In the experiment on the degree of recognition of traffic accidents, the performance of recognition of car wrecks and contacts was improved for all environmental information. These experiments show that V- TIDB can be used to learn traffic accident recognition models that take environmental information into account in detail and can be used for appropriate traffic accident analysis.

Type of Various Perspective Label

positive/negative (Traffic Incident)
Category (Incident Targets)
Contact_level (Contact Level)
Derivation_object (Derivation Object)
Environment (Environment)
Predictablility (Predictability)
Reaction (Reaction)
State (State)
Time (Time)
Traffic_lane_of_the_object (Traffic Lane)
Weather (Weather)

Description of the V-TIDB annotations json

Our annotation file is here.

database
- ID (format: {youtube_video_id}_{start_time}_{end_time})
  - annotations
    - labels (various-perspective labels assigned to the video in question)
    - segment
label (Key-value pairs for parent and child categories)

Citation

@InProceedings{VTIDB,
author = {Shota Nishiyama and Takuma Saito and Ryo Nakamura and Hirokatsu Kataoka and Kensho Hara},
title = {Traffic Incident Database with Multiple Labels Including Various Perspective Environmental Information},
booktitle = {Proc. IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
year = {2023},
}