Deaths from train strikes are still rising despite concerted efforts to lower them. According to the Federal Railroad Administration, trespassing incidents at the nation's 210,000 rail crossings result in hundreds of fatalities yearly in the United States. The FRA estimated in 2008 that 500 persons were killed each year when intruding on railroad rights-of-way. Ten years later, the FRA reported that the total number, including suicides, had increased to 855.
A research team has created a programme that uses AI to detect trespassing at train crossings and reduce fatalities, which have risen over the previous ten years. The development could propel the Artificial Intelligence As A Service (AIaaS) Market forward. Local law enforcement could place officers close to crossings during high violation rates. Further, railway owners and decision-makers could be advised of more efficient crossing solutions—such as grade crossing elimination systems or advanced signals and gates.
The newly developed framework with AI automatically detects railroad trespassing incidents. It also classifies different types of violators and produces videos of violations. The system employs an object detection algorithm to process video input into a single dataset. With this knowledge, one can respond to several queries, such as when people transgress the most and whether they circumvent the gates, going up or down.
The researchers obtained video footage shot at one crossing in urban New Jersey to put their theory to the test. Following 2015 Fixing America's Surface Transportation Act (FAST), cameras were installed at the study location. However, most crossing video systems today are either not evaluated or are examined manually, which is time-consuming and expensive.
One thousand six hundred thirty-two hours of historical video from the study site were used to train their AI and deep-learning technology. They found that 3,004 trespassing incidents—or an average of 44 per day—occurred throughout the 68 days of observation. The researchers also discovered that about 70% of trespassers were male, about a third did so before the train passed, and that Saturdays around 5 p.m. were the most common time for violations.
Through the risk analysis of their data feeds in specific places, researchers wish to provide the training industry and decision-makers with tools to harness the underutilised potential of video surveillance infrastructure.