Accident Video Fault Estimation Using Simulated 3D Learning Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for determining the percentage of fault in traffic accidents using video data from accident vehicles are inefficient due to the heavy workload in creating and maintaining databases, limited video data availability, and the difficulty in acquiring sufficient data for learning, making it difficult to achieve accurate fault determination.
Innovation Solution
An information processing device generates simulated moving images based on fault percentage rules, varying parameters such as 3D objects and conditions to create a large quantity of learning data, and uses these to develop a model for determining similarity with actual accident videos, enabling efficient and accurate fault percentage calculation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual creation and maintenance of accident case databases is performed, then fault determination can be conducted, but the workload is heavy and time-consuming
Solution Approach 1:
The patent creates a virtual copy of the accident scene by generating a simulated moving image that replicates the key elements (vehicles, pedestrians, road conditions, weather) from the actual accident video. This virtual model allows automated analysis without manual database creation, resolving the contradiction between accurate fault determination and time-consuming manual work.
Solution Approach 2:
The patent replaces the manual mechanical process of creating and maintaining accident databases with an automated information processing system. The system automatically extracts features from accident videos, generates simulated images, and performs fault analysis without human intervention, eliminating the time loss associated with manual database management.
2Extent of automation
If a model is created by learning from recorded accident video data, then fault determination can be automated, but sufficient video data is difficult to acquire
Solution Approach 1:
The patent generates synthetic copies of accident scenes through simulated moving images that replicate real accident conditions. These synthesized data copies expand the training dataset without requiring additional real accident videos, enabling automated model learning while overcoming the limitation of insufficient real-world data availability.
Solution Approach 2:
The patent varies parameters in the simulated moving images (such as vehicle positions, speeds, weather conditions, road layouts) to generate diverse training data from a limited set of real accident cases. This parameter variation creates multiple learning samples from single accident scenarios, resolving the contradiction between automation and data quantity.
3Measurement precision
If various accident situations are covered to create a comprehensive database, then fault determination accuracy improves, but the complexity of database creation increases
Solution Approach 1:
The patent segments the accident analysis into distinct components (vehicle extraction, pedestrian extraction, road condition analysis, weather condition analysis) and processes each separately. This segmentation allows comprehensive coverage of various accident situations without overwhelming complexity, as each component can be handled independently through automated feature extraction and simulated image generation.
Solution Approach 2:
The patent creates a universal simulated moving image generation system that can handle multiple types of accident situations (vehicle collisions, pedestrian accidents, various road conditions, different weather conditions) through a single integrated process. This multi-functional approach covers diverse accident scenarios without requiring separate database creation processes for each type, reducing overall complexity.
Data Source
AI summary
An information processing device (100) includes an acquisition unit (131) that acquires a fault percentage rule defined according to a situation at a time of occurrence of an accident and indicating a percentage of fault of a party of the accident, a moving image generation unit (132) that generates a simulated moving image assuming the situation at the time of occurrence of the accident which situation corresponds to the fault percentage rule, and a learning data generation unit (133) that generates learning data in which a learning moving image in which a component included in the simulated moving image is changed as a parameter is associated with the percentage of fault.