New AI Software Predicts Where the Next Car Accident is Most Likely to Happen

Developed by Israeli researchers, the software outputs optimal daily traffic police patrol schedules. Israeli police is planning to start testing the technology soon

CTECH by Calcalist / Yoav Stoler 1.01.18

Photo: Gil Nehushtan

Israel’s police force will soon be battling the country’s mounting road collision problem with a new weapon—an artificial intelligence-based crash prediction software developed by a team of researchers from Israel’s University Bar-Ilan.

Led by Sarit Kraus, head of Bar-Ilan’s computer science lab, and Ariel Rosenfeld, the team collected analytic data from car accidents in Israel over the past 13 years. Mapping Israel’s entire road network, the team took into account data such as the slope and angle of roads, the height of the sun and its effect on eyesight at the time of the accident.

“Our main challenge was to understand where the next accident will occur,” Mr. Rosenfeld said in an interview with Calcalist last week, adding that existing lists of dangerous roads used by Israeli police have proved inaccurate.

Rain, traffic congestion, public transportation strikes and incidents in progress must all be taken into account together with the road’s built-in risks, he explained. The output: suggested daily patrol schedules. Initially, the output will be used as recommendations only, Mr. Rosenfeld said.

The result is a software that uses around 120 variables to predict the probability of severe traffic accidents at every one of Israel’s routes, and suggest the placement of road police units accordingly within a 10 meter range of the predicted spot.