A leading team of engineers at Rutgers has innovated a railroad encroachment identification system valued at Artificial Intelligence. This development provides a significant solution to the rise in deaths at these sites that has been witnessed over the past decade..
The discovery of this remarkable invention was recently published in the journal Analysis and prevention of accidents.
Artificial intelligence in automated intrusion detection
The research team is made up of Asim Zaman, a project engineer at Rutgers, and Xiang Liu, a transportation professor at Rutgers. Both have worked on the development of a model assisted by Artificial Intelligence that autonomously identifies episodes of invasion of the railway system, in addition to differentiating between different types of invaders and generating video clips of each event. This intelligent model employs an object recognition algorithm to process video data into a single data set.
"This information will provide us with answers to various questions, such as the time with the highest number of invasions and whether people access the doors before getting on or off."Zaman mentioned.
In recent years, there has been an increase in accidents resulting from the violation of private property in the United States, causing the death of hundreds of people. Several approaches have been implemented to reduce these fatalities, however, none have proven to be effective to date.
In 2008, the FRA (Federal Railroad Administration) estimated the death of 500 people each year due to the invasion of train tracks, a number that rose to 855 in 2018 according to data from the same organization.
Zaman and Liu determined through their study that trespassers are unauthorized individuals or vehicles in a train zone or transit property not marked for public use, or those people who transit a marked zone after it has been activated.
Previous studies have primarily reported casualty information, without taking into account near misses, which Liu and Zaman say can offer valuable insight into invasion patterns, which could allow for the design of more effective control solutions.
The researchers' theory was examined with videos captured in an urban area of New Jersey. One of the difficulties with current video systems in these locations is that they are not systematically analyzed since this procedure is labor intensive and expensive.
Artificial intelligence training
Zaman and Liu trained their Artificial Intelligence y learning tool deep to analyze 1,632 hours of videos from the study site. After 68 days of follow-up, 3,004 cases of invasion were found, which represents an average of 44 each day. They found that almost 70% of the intruders were men and about a third agreed before the train passed. Most invasions occurred on Saturdays around 5 p.m.
According to Zaman, local authorities could make use of these detailed data to position police officers near the crossing during times of peak violations, or inform railway owners and decision makers about more effective crossing measures. These solutions could include grade crossing elimination systems or advance signs and gates.
“Everyone loves information and that is exactly what we provide. “We want to provide the rail industry and decision makers with tools to harness the latent potential of video surveillance infrastructure through risk analysis of their data sources at specific locations,” Liu added.
Researchers are also conducting studies in Virginia and North Carolina. They recently received a $583,000 US Department of Transportation grant to expand their analysis to other states, including Connecticut, Louisiana, and Massachusetts.
READ MORE ARTICLES ABOUT: Data Science with AI.
READ THE PREVIOUS POST: Neural networks learn most effectively by imitating human sleep patterns.