Vehicle Accident Data Feature Extraction and Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle systems face challenges in efficiently detecting and collecting accident-related driving experience event data, particularly in identifying rare events like accidents and near-accidents, as they generate vast amounts of irrelevant data, and existing methods struggle to adapt to new types of accidents or near-accidents without collisions.
Innovation Solution
A computer-implemented method that processes driving experience data streams to determine an accident similarity score and criticality score, storing relevant data features based on these scores, and transmitting them to a remote server, using machine learning models to identify both known and new types of accident or near-accident events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all driving experience data is transferred to remote server for analysis, then complete accident detection is achieved, but data transfer volume and processing burden increase significantly
Solution Approach 1:
The patent extracts and transfers only the essential accident-related features (extracted data features) rather than complete raw driving experience data. The vehicle system identifies accident events using local machine learning models, extracts relevant features from accident data segments, and transfers only these features to the remote server, significantly reducing data transfer volume while maintaining detection completeness.
Solution Approach 2:
The vehicle performs preliminary accident detection and feature extraction locally before transferring data to the remote server. The machine learning model on the vehicle pre-processes the data, identifies accident segments, extracts relevant features, and prepares the data for transfer, reducing the burden on the remote server and minimizing data transfer requirements.
2Measurement precision
If traditional collision-based accident detection methods are used, then clear accident events are detected, but near-accidents without collision are missed
Solution Approach 1:
The patent changes the detection parameters from binary collision detection to continuous similarity scoring. The machine learning model calculates an accident similarity score that reflects the degree of resemblance to known accident patterns, enabling detection of both clear accidents and near-accidents. This parameter change allows the system to identify events across a spectrum of severity rather than requiring definitive collision evidence.
Solution Approach 2:
The detection system transitions from static collision thresholds to dynamic similarity-based detection. The machine learning model adapts its detection criteria based on the calculated similarity scores, allowing flexible identification of various accident types including near-misses. The system dynamically adjusts what constitutes an detectable event based on the similarity assessment rather than fixed collision parameters.
3Measurement precision
If machine learning models use only historical accident data for training, then known accident types are detected accurately, but new accident types cannot be identified
Solution Approach 1:
The machine learning model is designed with multi-functionality to both classify known accident types and detect potential new accident types. The model uses historical accident data for training and validation while also evaluating new data segments against established patterns. When a data segment shows significant deviation from known patterns but still meets accident criteria, the system flags it as a potential new accident type, enabling the system to both classify known events and discover new ones.
Data Source
AI summary
A computer-implemented method and related aspects for detecting and collecting accident-related driving data on a vehicle are included. The method includes receiving a driving experience data stream from a source on-board the vehicle, determining an accident similarity score for a data segment of the driving experience data stream, wherein the accident similarity score is based on a similarity measurement of data features extracted from the data segment to data features of a historic accident or near-accident driving experience event, determining a criticality score for the data segment, and storing, in dependence on at least one of the determined accident similarity score and the determined criticality score, at least the extracted data features from the data segment as an accident-related driving experience event in association with at least the determined criticality score.