According to the Federal Railroad Administration (FRA) Office of Safety Analysis, there were 112 reportable accidents recorded in South Carolina in 2022. With more than 150,000 miles of railroad tracks in the U.S., trespassing is the top cause of all railroad-related deaths.
Since last September, Civil and Environmental Engineering Associate Professor Yu Qian and his research team have been working on a novel approach by combining geographic and demographic information to determine the reasons behind an individual’s dangerous decisions on or near railroad grade crossings.
Qian’s three-year, $555,000 project is funded by the FRA and Department of Transportation with partnerships from Mechanical Engineering Professor Yi Wang and Canadian company TRAINFO. The project, Discover Influential Geosocial Factors Aggravating Crossing Trespassing (DIGFACT), aims to connect community and railroad safety, a concept which has not been previously explored.
“The problem is not a lack of sufficient data, it's that we have too much data,” Qian says. “So how can we turn that data into information that helps improve railway safety to save lives?”
Qian aims to develop the first trespassing behavior analysis tool based on geographic and demographic information digitalization with a deep learning-enabled feature. The tool will determine the subtle correlation between trespassing and underlying social and community factors. It will also be capable of predicting trespassing probability or level of risks using location and the surrounding demographic conditions.
Qian’s team will analyze recorded trespassing events from FRA trespassing datasets. But it will also include related demographic information for determining the underlying factors that cause pedestrians or motorists to take risky actions. Qian previously worked with the FRA’s Human Factors research and development program, which oversees safety and accidents involving people.
We think demographics are the hidden factor to why people make these unsafe decisions.
- Yu Qian
While previous monitoring efforts were more passive preventive care, Qian’s project aims to actively understand the reasons why motorists or pedestrians decide to cross railroad tracks in risky situations.
“In terms of geographic factors, someone may cross the tracks if there’s a school, business district or residential area nearby. They do this because it’s close or may not be serious about railway safety,” Qian says. “But what about the demographics of population distribution, age group, and education and income levels? That also may contribute to pedestrians or motorists making unsafe decisions.”
Graduate student Yang Zhang is currently using all available datasets to collect geographic information such as street configurations and key business facilities (hospitals, schools). Additional information such as the location, fatalities, financial loss and specifics of each incident will be entered into a geographic information system (GIS) to show the location and frequency of accidents. Zhang admits that one of the challenges of collecting geographic information is managing the variability and inconsistency across different data sources.
“To address this challenge, we developed a flexible yet rigorous data integration workflow that allows us to harmonize diverse inputs while preserving essential geographic details,” Zhang says. “This workflow not only ensures the reliability of our analyses but also enables fast and targeted sampling of geographic features in regions of interest.”
Zhang’s data will be combined with accident and demographic and census data, including age, gender, education, and income distributions. Graduate student Yichuan Cao will compile this information. His work will include linking accident locations, which are captured as geographic coordinates, with defined statistical areas from U.S. Census data.
"We’ve selected the Census Block Group level as an example to demonstrate how these links can be established,” Cao says. “This integration allows us to examine variables such as population density, poverty levels, and housing patterns, and how they may interact with railroad infrastructure. The fine-grained analysis opens the door to designing more targeted and localized safety interventions, something that would be difficult to achieve using only broad-scale data."
The combined data will be digitized to create a map of the social and community profile.
“We've seen people use geographic information to predict highway congestion and accidents, but there's no demographics involved,” Qian says. “We think demographics are the hidden factor to why people make these unsafe decisions.”
Qian hopes to establish a link for creating a graphic neural network to determine a correlation to identify the most sensitive factors. To facilitate artificial intelligence development, Qian’s team will formulate trespassing evaluation and community characteristics. Advanced deep learning techniques will be used to find complex patterns hidden in pedestrian and motorist incursions on railroad tracks.
If sensitive factors are identified, Qian will map the data into different cities and identify hot spots where the FRA needs to monitor and allocate resources. This will affect future development and hopefully lead to eliminating or reducing future incidents.
TRAINFO, a company exclusively dedicated to eliminating traffic delays and accidents at rail crossings, is the project’s industry partner. Qian plans to validate his technology developed through DIGFACT to TRAINFO, who have already installed their safety systems in several U.S. metropolitan areas. TRAINFO is providing a $100,000 in-kind support and once the tool is developed, Qian will select a major U.S. city for installation.
“They will install all the equipment and attempt to validate or provide information to improve our model to see if it works or not. If it works, it will be validated for the FRA to make future decisions,” Qian says.
The success of the project will significantly enhance the understanding of why people would rather risk their lives by trespassing on railroad tracks and provide educational information and countermeasures for improving the quality of life.
“We’re hoping to provide a list of all the sensitive and dominant factors. This list and our models will predict where the hotspots are in particular cities while trying to prevent contributing factors happening together,” Qian says. “The primary goal is to safely reduce the number of accidents and save people's lives.”