Citation

Li Y, Karim MM, Qin R, Sun Z, Wang Z, Yin Z. Accid. Anal. Prev. 2020; 151: e105962.

Copyright

(Copyright © 2020, Elsevier Publishing)

DOI

10.1016/j.aap.2020.105962

PMID

unavailable

Abstract

Reducing traffic fatal crashes has been an important mission of transportation. With the rapid development of sensor and Artificial Intelligence (AI) technologies, the computer vision (CV)-based crash anticipation in the near-crash phase is receiving growing attention. The ability to perceive fatal crash risks in an early stage is of paramount importance as well because it can improve the reliability of crash anticipation. Yet this task is challenging because it requires establishing a relationship between the driving scene information that CV can recognize and the fatal crash features that CV will not get until the crash occurrence. Image data with the annotation for directly training a reliable AI model for the early visual perception of fatal crash risks are not abundant. The Fatality Analysis Reporting System (FARS) contains big data on fatal crashes, which is a reliable data source for finding fatal crash clusters and discovering their distribution patterns to tell the association between driving scene characteristics and fatal crash features. To enhance CV's ability to perceive fatal crash risks earlier, this paper develops a data analytics model from fatal crash report data, which is named scenario-wise, spatio-temporal attention guidance. First, the paper identifies five descriptive variables that are sparse and thus allow for decomposing the 5-year (2013-2017) fatal crash dataset to develop scenario-wise attention guidance. Then, an exploratory analysis of location- and time-related descriptive variables suggests dividing fatal crashes into spatially defined groups. A group's temporal distribution pattern is an indicator of the similarity of fatal crashes in the group. Hierarchical clustering and K-means clustering further merge the spatially defined groups into six clusters according to the similarity of their temporal patterns. After that, association rule mining discovers the statistical relationship between the temporal information of driving scenes with fatal crash features, such as the first harmful event and the manner of collisions, for each cluster. The paper illustrates how the developed attention guidance supports the design and implementation of a preliminary CV model that can identify agents of a possibility to involve in fatal crashes from their environmental and context information.

Language: en

Keywords

Visual perception; Clustering; Association rule mining; Attention guidance; Crash report data; Spatio-temporal information