Analyzing Relationship Between Head-on Collision and Road Curvature

This presentation presents an analysis of the relationship between head-on collision and road curvature. This study was published at IIAI AAI 2025.

Transcript

1. Analyzing Relationship Between Head-on Collision and Road Curvature IIAI AAI 2025 18th International Congress on Advanced Applied Informatics Keisuke ANDO1 Takeshi UCHITANE1 Naoto MUKAI2 Kazunori IWATA3 Nobuhiro ITO1 1Aichi Institute of Technology 2Sugiyama Jogakuen University 3Aichi University 17th International Conference on Smart Computing and Artificial Intelligence Kitakyusyu, Japan, 13-19 July 2025
2. THE CRITICAL IMPACT OF HEAD-ON COLLISIONS Fatal Accidents by Type Crosses the Center Line Head-on collisions are very dangerous High relative speed Most frequent type 2023 2,618 cases Head-on collisions 829 cases Japan’s National Police Agency 2
3. OBJECTIVE Curved Roads To clarify the relationship between Head-on Collisions Insights from Japanese Traffic Accident Data Preventing head-on collisions with a high fatality risk and 3
4. RELATED WORKS Road curvature affects the gaze points Michael and Benjamin (2001), Osugi et al. (2021) Changes in road curvature patterns are cause accidents Akahane et al. (2022) Road curvature impacts the occurrence accidents What about the relationship with head-on collisions? 4
5. PRELIMINARY INVESTIGATION Kitakyushu Aichi: high number of accidents Aichi Prefectural Police We Accident data: 540,000 traffic accidents (2008-2020) 5
6. AREAS PRONE TO HEAD-ON COLLISIONS Mountainous areas Coastal areas Proportion of head-on collisions (%): Low High 3-5 5-10 10+ 1-3 0-1 Honshu Pacific Ocean Area with a high proportions 6
7. ROADS OF HEAD-ON COLLISION HOTSPOTS Accident count heatmap (Toyota): 8 collisions have occurred on a winding road 500 m road accident occurred 7
8. PRELIMINARY FINDINGS • High proportion of head-on collisions are observed in mountainous and coastal areas. • Head-on collisions are concentrated on winding roads. Does road curvature contribute to head-on collisions? ? Key Question: 8
9. Analyzing accidents on curved roads to determine the impact of road curvature on head-on collisions APPROARCH AND METHODS 1. Data preparation: calculate road curvatures 2. Statistical testing of road curvature: “Is road curvature higher at locations with head-on collisions?” 3. Decision tree analysis of head-on collisions: “How much do road curvature affects head-on collisions?” 9
10. ACCIDENT DATA Number of head-on collisions: Aichi: 5,932 accidents (2008-2020) Number of samples is limited Japan: 11,688 accidents (2019-2023) The analysis uses nationwide traffic accident data Number of samples is relatively large 10
11. FEATURES 1. Time periods morning, daylight, evening, … 2. Weather clear, cloudy, rainy, … 3. Area type urban or non-urban 4. Median strip guardrail median, center line, … 5. Pedestrian separation fences, curbs, paved shoulder, … 6. Day of the week Monday, Tuesday, … 7. Holiday status holiday, previous day, or other day 8. First party’s age 0-24, 25-34, 35-44, 45-54, … 9. First party’s vehicle type passenger car or cargo vehicle 11 10 features are contained in data: 10. Road curvature and
12. CALCUALTION OF ROAD CURVATURE Circumcircle Points at a fixed distance from The analysis uses OpenStreetMap for road data Calculating curvature using Menger curvature This method can calculate curvature from line segments OpenStreetMap doesn’t provide road curvature 12
13. STATISTICAL TEST OF ROAD CURVATURES All accident data Head-on collisions Others Separate Data shows difference Data shows no difference Road curvature is higher at locations with head-on collisions High curvature High curvature 13
14. ANALYSIS OF FACTOR USING DECISION TREE Factor Part of the decision tree: How much road curvature influences head-on collisions? TRUE FALSE 14 Any of 10 features Generated by the CART algorithm: classifies data based on Gini impurity See more: https://github.com/ NONONOexe/curve-headon- collision-analysis Prone to other collisions Prone to head-on collisions
15. HEAD-ON COLLISION DATA Accidents 31,237 records Head-on collisions Others 11,688 records (37.4 %) 19,559 records (62.6 %) Separate • On curves • On OpenStreetMap roads • National accidents Compare 15
16. STATISTICAL TEST RESULTS Probability Densities 9 6 3 0 0.00 0.05 0.10 0.15 Others Head-on collisions Statical analysis confirmed that head-on collisions group had significantly greater road curvature (𝛼 = 0.01) There is a difference in the medians 16
17. DECISION TREE FOR ACCIDENT FACTORS Consistent with preliminary: mountainous / coastal areas Median strip Area type Road curvature [1/m] Center line or None Non-urban Guardrail or wire rope Urban 0.0104 ≦ < 0.0104 Prone to head-on collisions Prone to other collisions 17
18. ROAD GEOMETRY OF HIGH-RISK LOCATIONS Accident point Driver’s location Road curvature greater than 0.0104 [1/m] 3DCG image This road is difficult to see and dangerous for driver! 18 Can we turn safely?
19. CONCLUSION Road curvature contributes to head-on collisions • Road curvature is significantly higher at locations where head-on collisions occur • The risk of head-on collisions increases when road curvature exceeds 0.0104 [1/m] • Road curvature holds relatively high feature importance for head-on collisions 19
20. SUMMARY • Objective: To clarify the relationship between head-on collisions and road curvature • Preliminary: Revealed head-on collisions are tended to occur on winding roads in Aichi Prefecture • Investigated the relationship using nationwide Japanese data • Statistical tests confirmed that road curvature is significantly greater at head-on collision locations • Decision tree analysis confirmed that road curvature increases the risk of head-on collisions 20