A reproducible approach to merging behavior analysis based on High Definition Map

Citation

1
2
3
4
5
6
7
8
9
10
@article{
Yang Li, 2023
"https://arxiv.org/abs/2303.11531"
title={A reproducible approach to merging behavior analysis based on High Definition Map},
author={Yang Li and Yang Liu and Daiheng Ni and Ang Ji and Linbo Li and Yajie Zou},
year={2023},
eprint={2303.11531},
archivePrefix={arXiv},
primaryClass={cs.RO}
}

Code Structure

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
A reproducible approach to merging behavior analysis based on High Definition Map/
|__ asset: analysis results
| |__ acceptedGaps
| |__ backgroundPicture
| |__ consecutiveLaneChange
| |__ JSDivergence
| |__ preliminaryAnalysis
| |__ safetyAnalysis
| |__ table
| |__ trafficFlowSpeed
|__ conf: experiment configurations
| |__ lanelet2map
|__ drone-dataset-tools-master: dataset
|__ src: merging trajectories extraction, indicators calculation, results visulization
| |__ extraction
| |__ figure
| | |__accepetedGapAnalysis
| | |__consecutiveLanechangeAnalysis
| | |__preliminaryAnalysis
| | |__safetyAnalysis
| | |__similarityAnalysis
| | |__trafficFlowSpeedDensity
| |__ table
|__ utils: utility functions
|__ requirements.txt: required packages

Abstract:
Existing research on merging behavior generally prioritize the application of various algorithms, but often overlooks the fine-grained process and analysis of trajectories. This leads to the neglect of surrounding vehicle matching, the opaqueness of indicators definition, and reproducible crisis. To address these gaps, this paper presents a reproducible approach to merging behavior analysis. Specifically, we outline the causes of subjectivity and irreproducibility in existing studies. Thereafter, we employ lanelet2 High Definition (HD) map to construct a reproducible framework, that minimizes subjectivities, defines standardized indicators, identifies alongside vehicles, and divides scenarios. A comparative macroscopic and microscopic analysis is subsequently conducted. More importantly, this paper adheres to the Reproducible Research concept, providing all the source codes and reproduction instructions. Our results demonstrate that although scenarios with alongside vehicles occur in less than 6% of cases, their characteristics are significantly different from others, and these scenarios are often accompanied by high risk. This paper refines the understanding of merging behavior, raises awareness of reproducible studies, and serves as a watershed moment.

Keywords:
Merging Behavior, High Definition Map, Reproducible Research, exiD dataset.

Research bullets

1
2
3
4
5
6
7
8
9
10
11
12
13
14
1. Merging distance
2. Merging distance ratio
3. Merging duration
4. Merging lateral trajectory
5. Merging lateral speed
6. Scenario classification
7. Alongside vehicle identification
8. HD map pre-labeling
9. Standardized indicators
10. Safety analysis
11. Traffic flow-density-speed
12. Accepted gap analysis
13. Scenario similarity
14. Consecutive lane-changing duration