Today’s world is one big maze connected by layers of concrete and asphalt that provides us with the luxury of car navigation. For many road advancements — GPS allows us to fire fewer neurons thanks to map apps, cameras alert us to expensive dents and scrapes, and self-driving electric cars have lower fuel costs — safety measures haven’t worked our own. We still depend on a steady diet of traffic lights, confidence, and the steel around us to get safely from point A to point B.
To overcome the uncertainty inherent in collisions, scientists from the Computer Science and Artificial Intelligence Laboratory (CSAIL) of the Massachusetts Institute of Technology and the Qatar Center for Artificial Intelligence have developed a deep-learning model that predicts with high-resolution collision risk maps. Equipped with a combination of historical collision data, road maps, satellite imagery and GPS traces, the hazard maps describe the expected number of accidents over a period of time in the future to identify high-risk areas and predict future accidents.
These types of hazard maps are usually captured at a much lower resolution hovering around hundreds of metres, which means important details are hidden as the roads become blurred together. These maps, though, are 5 by 5 meter grid cells, and the higher resolution brings new clarity: Scientists have found that a highway, for example, has higher risks than nearby residential roads, and slopes that merge and exit the road Fast has higher risks. Danger than other methods.
“By defining a basic risk distribution that determines the probability of future accidents in all places, and without any historical data, we can find safer routes, enable auto insurers to offer customized insurance plans based on customers’ driving routes, and help city planners design Safer roads, and even anticipate future accidents,” says MIT CSAIL Ph.D. Student Songtao He, lead author on a new paper on the research.
Although car accidents are rare, they cost about 3 percent of global GDP and are the leading cause of death among children and young adults. This discrepancy makes deducing maps with such high accuracy a difficult task. Faults at this level are sparsely scattered – the average annual odds of a 5 x 5 grid cell failing is one in 1,000 – and they rarely occur at the same location twice. Previous attempts to predict collision risk have been largely “historic”, as an area would only be considered high-risk if there was a previous collapse nearby.
The team’s approach casts a broader net to capture critical data. Identifies high-risk locations using GPS trajectory patterns, which provide information about traffic density, speed, and direction, and satellite imagery describing road structures, such as the number of lanes, whether there is a shoulder, or if there are too many pedestrians . Then, even if there is no high-risk area with recorded incidents, it can still be identified as high-risk, based on traffic patterns and topology alone.
To evaluate the model, the scientists used accidents and data from 2017 and 2018, and tested its performance in predicting 2019 and 2020 accidents. Several sites were identified as high-risk, although there were no recorded incidents, and also experienced accidents during the follow-up years.
Amin Sadeghi, Principal Scientist at Qatar Computing says, “Our model can generalize from one city to another by combining multiple clues from seemingly unrelated data sources. This is a step towards AI, because our model can predict fault maps in unknown areas.” . Research Institute (QCRI) is an author on the paper. “The model can be used to infer a useful crash map even in the absence of historical collision data, which can translate into positive use for city planning and policy making by comparing imaginary scenarios.”
The dataset covered 7,500 square kilometers of Los Angeles, New York, Chicago and Boston. Of the four cities, Los Angeles was the most dangerous, with the highest crash density, followed by New York City, Chicago, and Boston.
“If people can use Hazard Map to identify potentially high-risk parts of the road, they can take action in advance to reduce the risk of their trips. Apps like Waze and Apple Maps have accident feature tools, but we try to get before collisions — before they happen. ‘, he says.
The likelihood of fatal car accidents increases during the pandemic
Infer high-resolution traffic accident risk maps based on satellite imagery and GPS tracks. openaccess.thecvf.com/content/… _ICCV_2021_paper.pdf
Provided by the Massachusetts Institute of Technology
This story is republished with permission from MIT News (web.mit.edu/newsoffice/), a popular site covering news about MIT research, innovation, and education.
the quote: Deep learning helps predict traffic accidents before they happen (2021, Oct 13) Retrieved Oct 13, 2021 from https://techxplore.com/news/2021-10-deep-traffic.html
This document is subject to copyright. Notwithstanding any fair dealing for the purpose of private study or research, no part may be reproduced without written permission. The content is provided for informational purposes only.