Driving Scene-Based Hazard Prediction With Selective AI Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computing devices used for driving have limited resources, making it difficult to perform highly accurate dangerous scene prediction due to the resource constraints.
Innovation Solution
A dangerous scene prediction device that selects and uses a suitable learning model based on the driving scene, reducing computation load by associating specific learning models with different driving scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all dangerous scenes are predicted using comprehensive learning models, then prediction coverage is improved, but computation load increases beyond available resources
Solution Approach 1:
The patent divides the dangerous scene prediction task into multiple segments by creating separate learning models for different driving scenes (e.g., urban areas, highways, parking lots). Each learning model is specialized for specific scene types, allowing the system to process only relevant predictions for the current driving context, thereby reducing overall computation load while maintaining comprehensive coverage across all scene types.
Solution Approach 2:
The system dynamically selects which learning model to execute based on the current driving scene classification. Rather than running all prediction models simultaneously, the system adapts its computation by activating only the appropriate model for the detected scene type, optimizing resource utilization while maintaining prediction accuracy for relevant dangerous scenes.
2Measurement precision
If comprehensive learning models are used for all dangerous scenes, then prediction accuracy is improved, but resource consumption exceeds available computing resources
Solution Approach 1:
The patent applies local quality by tailoring the complexity and type of learning model to match the specific requirements of each driving scene. For example, simpler models may be used for low-risk scenes while more sophisticated models are deployed for high-risk scenarios. This ensures that computational resources are concentrated where they provide the most value, maintaining high prediction accuracy for critical scenes while conserving overall resource consumption.
3Reliability
If multiple learning models are executed simultaneously for all dangerous scenes, then prediction completeness is improved, but processing time increases beyond acceptable limits
Solution Approach 1:
The patent segments the prediction processing into multiple specialized models that can be selectively executed. By dividing the comprehensive prediction task into scene-specific sub-tasks, the system processes only the necessary predictions for the current driving context, significantly reducing processing time while maintaining completeness for relevant dangerous scenes.
Solution Approach 2:
The system dynamically adjusts its processing strategy by selecting and executing only the learning model appropriate for the current scene type. This dynamic selection eliminates the time penalty of running unnecessary models while ensuring that the correct model is always active for the given driving conditions, thereby maintaining prediction completeness without excessive processing time.
Data Source
AI summary
A dangerous scene prediction device 80 for predicting a dangerous scene occurring during driving of a vehicle includes a learning model selection/synthesis unit 81 and a dangerous scene prediction unit 82. The learning model selection/synthesis unit 81 selects, from two or more learning models, a learning model used for predicting the dangerous scene, depending on a scene determined based on information obtained during the driving of the vehicle. The dangerous scene prediction unit 82 predicts the dangerous scene occurring during the driving of the vehicle, using the selected learning model.