Road Surface Recognition Using Dual Models Across Multiple Road Types
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition systems for vehicles require changing trained models based on the type of road being traveled, which is inefficient and labor-intensive.
Innovation Solution
Utilizing a combination of a first trained model generated by unsupervised training to identify areas with the same road surface attribute and a second trained model generated by supervised training to identify boundaries, allowing for the recognition of travelable areas without changing models based on road type.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a trained model is selected based on the type of road the vehicle travels on, then the recognition accuracy of the travelable area is improved, but the system complexity and operational effort increase due to the need to switch between multiple trained models
Solution Approach 1:
The patent segments the road surface recognition task into two independent components: (1) area segmentation using a first trained model to divide the road surface into multiple areas with different attributes, and (2) boundary detection using a second trained model to identify boundaries between these areas. This segmentation allows the system to maintain high recognition accuracy for different road types without requiring complete model switches, as each component can operate independently and consistently across various road conditions.
Solution Approach 2:
The patent creates a universal system that can handle multiple road types (roadways, sidewalks, bicycle paths, etc.) using the same combination of trained models. The first trained model performs area segmentation universally, and the second trained model detects boundaries universally, eliminating the need to switch between different complete trained models for different road types. This multi-functional approach maintains accuracy while reducing system complexity.
2Measurement precision
If multiple trained models are maintained for different road types, then the recognition accuracy for various road conditions is improved, but the time and effort required to change models increases
Solution Approach 1:
The patent performs preliminary action by pre-segmenting the road surface into areas with different attributes using the first trained model before boundary detection. This preliminary area segmentation creates a structured foundation that allows the second trained model to efficiently detect boundaries without requiring model switching. The preliminary classification of road surface areas enables rapid adaptation to different road types without the time-consuming process of changing complete trained models.
3Device complexity
If a single trained model is used for all road types, then the system complexity is reduced, but the recognition accuracy for specific road conditions deteriorates
Solution Approach 1:
The patent segments the recognition task into two specialized components: area segmentation (first trained model) and boundary detection (second trained model). This segmentation allows each component to be optimized for its specific function while working together to achieve high overall accuracy. The first model focuses on dividing the road surface into meaningful areas, while the second model focuses on detecting boundaries, creating a division of labor that improves accuracy without requiring a completely different model for each road type.
Solution Approach 2:
The patent adds a dimensional aspect to the recognition system by separating the analysis into two dimensions: area-level analysis (first trained model outputting multiple areas with attributes) and boundary-level analysis (second trained model outputting boundaries between areas). This dimensional separation allows the system to capture both the overall road surface characteristics and the precise boundary locations, achieving high accuracy while maintaining a unified system structure.
Data Source
AI summary
Provided is an image recognition device configured to execute using a first trained model, which is trained to receive input of an image including a road surface to output areas estimated to have road surfaces with the same attribute in the image as one area, and a second trained model, which is trained to receive input of an image including the road surface as an input to output boundaries of a plurality of the areas in the image, and recognizing a travelable area in which a moving body is capable of traveling on the basis of the areas and the boundaries obtained by inputting an image including a road surface captured by a camera mounted in the moving body to the first trained model and the second trained model.