In cooperation with the University of Massachusetts, Qatari artificial intelligence predicts accidents before they happen

The dataset collected by the task force covered 7,500 square kilometers from Los Angeles, New York, Chicago and Boston. Of the four cities, Los Angeles was the most dangerous, with the highest accident rate compared to other cities.

Artificial intelligence

Our world today is a large maze connected by layers followed by layers of asphalt concrete, which gives us the luxury of moving our vehicles and cars in the middle of this maze to reach our offices and homes on the sides of the roads.

For many technical developments related to the roads, the applications of maps and the Global Positioning System (GPS) allow us the luxury of navigating this maze without fear of getting lost, there are cameras, radar stations, electric cars that spend less fuel, and self-driving vehicles, As well as traffic lights that regulate traffic.

However, fatal accidents still occur every day, as more than 1.3 million people lose their lives every year due to road accidents, and every year between 20 and 50 million injuries are recorded due to these accidents, many of which end with permanent disabilities, according to United Nations statistics . finally published.

Injuries caused by road accidents also cause significant economic losses to individuals, their families and the countries as a whole, and these losses result from the cost of treatment and the loss of productivity for those who were killed, injured, or permanently disabled as a result of the accidents they were exposed to, in addition to family members who need leave from work or The school for the care of the injured.

Risk maps

To overcome this global conundrum, road insecurity and the uncertainty inherent in accidents, scientists from MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) and the Qatar Center for Artificial Intelligence have developed QCAI is a deep learning model that predicts possible accidents with high accuracy, and draws highly accurate maps that identifies the danger points on the streets and roads on which accidents will occur and when they will occur.

Scientists from the Computer Science and Artificial Intelligence Laboratory (CSAIL) at the Massachusetts Institute of Technology (MIT) have developed it for indoor use.New deep learning model predicts possible accidents with high accuracy (MIT)

These hazard maps are provided with a set of data and information about collisions and accidents that have occurred in the past on these roads, in addition to satellite images, and data from the Global Positioning System (GPS), all of which describe the expected number of accidents over a certain period of time. With the aim of identifying high-risk areas and predicting future incidents, as stated in the research that was recently published on the University of Massachusetts platform .

Prediction accuracy of 5 meters

The accuracy of the previous hazard maps was much lower, at best reaching hundreds of metres, but with the new model, the accuracy of accident prediction is only about 5 metres, which means extremely accurate prediction of accidents.

Scientists at the two institutes also found that highways contain higher risks than residential roads, and they also found that secondary roads that enter or exit highways have much higher risks than secondary roads connected to other roads.

“By defining a basic risk distribution table that determines the probability of future accidents occurring in all roads and places, and without any prior historical data, we can find safer routes and enable Auto insurers can offer customized insurance plans based on customers’ driving routes, as well as help city planners design future roads and streets that are safer, and even predict future accidents.”

Routes as grid cells

Car accidents cost about 3% of global GDP and are the leading cause of death for children and young people in the world. Which makes such work very important. In fact, mapping risks with such high accuracy was a very difficult task. The team mapped the roads as grid cells of 5 meters by 5 meters, and the probability of an accident in each cell consisting of 5 square meters, amounting to about one to a thousand, and the accident rarely occurs in the same location twice, which made the prediction process very difficult without historical precedents, as the area will not be considered high-risk unless accidents occur repeatedly.

The team's approach was to build a broader network to capture critical data that identifies high-risk locations using GPS traffic patterns, which provides information on traffic density, speed, and direction, as well as satellite images that describe road structures, such as the number of lanes in each a road, or whether there are a large number of pedestrians, or sub-roads entering or exiting it, and then, even if there is no high-risk area with previously recorded accidents, it can still be identified as high-risk based on traffic patterns Traffic and its topology.

To evaluate the model, the scientists used accidents and data from 2017 and 2018, and tested its performance in predicting accidents in 2019 and 2020, where many sites were identified as high-risk, although there were no previously recorded accidents, and accidents that occurred over years were followed. studying.

Qatar Artificial Intelligence: Our model can be applied in any city in the world

Researcher Amin Sadeghi, a scientist at the Qatar Center for Artificial Intelligence and a co-author of the research, says, “Our model can be generalized and applied in any city or place in the world by combining multiple evidence from data sources that may seem uncorrelated with each other.. This is a step towards artificial intelligence.” General because our model can predict accident maps in any region of the world, even in the absence of records of previous accidents, which can be used - if available - for city planning and traffic policy-making by comparing imaginary scenarios that could occur in the future.

The dataset collected by the task force 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 accident rate compared to other cities, followed by New York, Chicago and Boston, respectively.

Songtao He stresses: People can use our maps to identify potentially high-risk parts of the road, so they can plan their trips in advance to avoid these places, and apps such as Waze and Apple Maps have special tools to locate accidents, but in our model and maps we try to avoid accidents before they even happen.