News

Top > Press Release > GeoTechnologies Develops Intersection Risk Estimation Model Based on People Flow Data
2024
May
30

GeoTechnologies Develops Intersection Risk Estimation Model Based on People Flow Data

AI calculates the risk of traffic accidents at daily accessible intersections and makes it possible to ascertain the risk level.

  • X(Twitter)
  • Facebook
  • LinkedIn

GeoTechnologies, Inc. (headquartered in Bunkyo-ku, Tokyo; Hiroshige Sugihara, Chairman and Representative Director; Yoichiro Yatsurugi, President and CEO; hereinafter “GeoTechnologies”), an ESG metaverse company, has successfully developed an intersection risk estimation model that calculates accident risk at familiar intersections such as those in residential areas by combining our people flow data with AI technology.

With this development, it is now possible to calculate accident risks at small intersections consisting of residential roads, which have been considered difficult to calculate in the past. This makes it possible to identify in advance dangerous intersections hidden in daily accessible roads that people usually use for commuting to work or school.

By identifying dangerous intersections, the system can be expected to contribute to accident prevention efforts, such as presenting safer transportation routes that avoid the intersections in question, alerting drivers and residents in the vicinity, and improving road maintenance to enhance safety and security.

Social Background of Development

According to statistics from the Ministry of Land, Infrastructure, Transport and Tourism, the number of traffic fatalities in Japan has been steadily decreasing since 1995*, reaching a record low of 2,610 in 2022. On the other hand, the decrease in the number of traffic accidents on residential roads has been very small, with one out of every four accidents occurring on major roads in 2022. Therefore, the most important issue for further reduction of traffic accidents in the future is how to reduce the number of traffic accidents on roads used for daily life.

*Ministry of Land, Infrastructure, Transport and Tourism, Traffic Accidents

Features of the Intersection Risk Estimation Model

The most distinctive feature of the intersection risk estimation model we have developed is that it identifies traffic patterns (straight, right and left turns) at intersections by generating all users’ traffic trajectories based on human flow data , and incorporates cross-traffic and merge-traffic as risk estimation factors.

For example, straight traffic facing each other is unlikely to come into contact with each other as shown in the blue figure, while cross traffic and merge traffic may come into contact with each other as shown in the red figure. Thus, by incorporating the estimation factors of cross-traffic and merge-traffic, we can make more effective estimation than by simply estimating traffic volume.

cross-traffic and merge-traffic

Since our people flow data includes information not only on vehicles but also on pedestrians, it is possible to create pedestrian-specific traffic trajectories. By utilizing the characteristics of pedestrians, it is possible to estimate the risk of accidents between vehicles and pedestrians, in addition to estimating the risk of accidents between vehicles.

Outline of Research and Development

Based on the characteristics of each intersection, such as the volume of vehicle and pedestrian traffic and the structure of the intersection, taking into account traffic patterns calculated from our human flow data, and using traffic accident information* published by the National Police Agency as the correct data, we created a model to estimate traffic accident risk at intersections. Accuracy was verified by comparing the accident information with the results of risk estimation in a specific area.

*Open data of traffic accident statistics, National Police Agency

Use Data

The following data are used in each model constructed in this study.

Traffic Accident Information
Period 4 years from January 2019 to December 2022
subject Accidents between vehicles, and between vehicles and pedestrians
People flow data
Period 1 month from June 1, 2022 to June 30, 2022
Surveyed subject

The risk of accidents between vehicles and between vehicles and pedestrians was estimated and evaluated for each target area.

subject Unsignalized intersections consisting of residential roads
Target Area 4 areas Urban area (more/less accidents), rural urban area (more/less accidents) *
*Calculated based on statistical tables from the National Police Agency
Results of the experiment

Among the results of risk estimation, the points that exceeded a certain risk value were set as dangerous intersections, and their accuracy was verified by comparing them with intersections where accidents actually occurred based on traffic accident statistics.

As a result, it was confirmed that although the number of intersections identified as dangerous was 115, a very small number compared to the total of 13,787, the actual number of accidents among these intersections was 70, a high percentage that was consistent with the total number of accidents that had occurred.

This means that we were able to identify with a high degree of accuracy the dangerous intersections where accidents had actually occurred, and also that the 45 intersections we missed were potentially dangerous intersections with similar characteristics to those where accidents had occurred, although no accidents had occurred.

Relationship Between High-Risk Intersections and Accident Information

Breakdown of intersections judged to be dangerous

Contribution of features in intersection risk estimation model

In the risk estimation of each type of accident in the intersection risk estimation model, the contributions of features were extracted in order of magnitude, with cross-traffic and merging traffic contributing a high percentage in both vehicle-to-vehicle accidents and vehicle-to-pedestrian accidents. In addition, pedestrian-related features also contributed a high percentage to the risk estimation for both vehicles and pedestrians.

These indicate that among the extracted features, cross-traffic and merging traffic, in particular, contribute significantly to the accuracy of risk estimation. Furthermore, when estimating the risk of accidents between vehicles and pedestrians, the feature that contributes the most is pedestrian cross-traffic, which means that pedestrian information is an essential and critical element for improving the accuracy of the estimation.

contribution

Future Development

The results of the study confirm that the constructed intersection risk estimation model is capable of estimating accident risk with high accuracy.

This time, we built a model to estimate the universal risk of an intersection without considering the time of day. However, the amount of traffic at an intersection change from moment to moment depending on the time of day, such as morning, noon, and evening, and the risk is expected to change accordingly. Since our human flow data contains time information, we are working hard to develop a model to estimate the risk at each time of day.

We will also work toward commercialization by expanding the scope of target intersections, building a model to enable risk estimation including intersections with traffic signals, and then extending the area to the entire country for evaluation.

Contact Us

Companies wishing to utilize this “Intersection Risk Estimation Model” should contact us at the URL below.

About GeoTechnologies

Since our founding in 1994, we have consistently provided reliable digital maps. The following year, we launched a map software, "MapFan." Since then, we have been at the forefront of the industry by swiftly providing essential AD/ADAS maps, crucial for advanced autonomous driving, along with Japan's first i-mode maps, car navigation systems, corporate map data, and location-based solutions. Moreover, through mobile applications like "TRIMA" (2020), we have established real-time connections with our users, allowing us to grasp "insights" into real-world situations, such as human movement and underlying motivations. By combining vast amounts of large datasets, most notably people flow trends with spatial data accumulated over approximately 30 years, we utilize cutting-edge technologies that analyze and provide "insights of the moment." Our goal is to contribute to society by fostering a more comfortable and sustainable world.

Headquarters' location: 22F Bunkyo Green Court Center Office, 2-28-8 Honkomagome, Bunkyo-ku, Tokyo
Representative:
Yoichiro Yatsurugi, President & CEO
Founded: May 1, 1994
Business domains: Automotive business
Enterprise business
Marketing business
Consumer business