China’s Baidu Creates AI To Warn of Crowd Stampedes

This post is also available in: עברית (Hebrew)

In New Year’s Eve of 2015, thousands in Shanghai came to the Bund where a light show was to take place to mark the new year. A dangerously dense crowd formed, and at a viewing platform overlooking the river, confusion and a lack of crowd control led to panic and eventually stampede. Within minutes, 36 people had been killed and another 49 had been injured in the crush.

This might a memorable one, but it is by no means the only stampede in China’s collective memory. The number of people cause masses in the country to be, well, massive. In order to tackle this issue, Baidu (China’s largest research engine)’s Big Data Lab has offered a solution: data-crunching predictive A.I.

A paper presenting the system describes how it uses Baidu Maps data to predict areas where dangerous crowds could be forming, and warn both users and authorities in advance. Baidu researchers figured that areas with spiking Baidu Map queries from users at nearby locations might be places where crowds are likely to form in the near future, as many consult the digital map before choosing a route on their way to somewhere. They studied mapping data from a number of high-density crowd events, including the 2014 stampede, and found that this was indeed the case. The team devised a method to crunch Baidu’s mapping query data in real time and output a warning if the number of queries on a single area from nearby locations crossed a specific threshold. Then, a machine learning system was implemented to help crunch historical data and make accurate predictions about future danger areas.

Although the system hasn’t been implemented anywhere yet, real-world implementation is the ultimate goal. “We believe the successful deployment of our method can bring many benefits to our society,” the paper concludes.