Vehicle External Recognition Using Feature Point Distance and Image Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle external recognition systems struggle to accurately identify road surface areas far from the driver's vehicle, where feature points are scarce and edge observation is difficult, making it challenging to measure three-dimensional information and extract road surfaces effectively.
Innovation Solution
An on-vehicle external recognition apparatus that includes an image acquisition unit, feature point distance measurement unit, first road surface area extraction unit, image feature amount calculation unit, and travelable area recognition unit, which measures distances to feature points, calculates multi-dimensional image features, and recognizes travelable areas by integrating first and second road surface areas extracted based on similarity and camera geometry.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature point tracking is used to measure three-dimensional information, then measurement precision is improved in areas with strong edges, but measurement capability deteriorates in road surface areas far from the vehicle where feature points are scarce
Solution Approach 1:
The patent introduces an intermediary approach by using feature points as mediators to infer road surface areas. Instead of directly detecting road surfaces in distant areas where feature points are scarce, the system uses feature points in visible areas to calculate distances and infer the positions and extents of road surfaces that are harder to observe directly. This intermediary method bridges the gap between observable feature points and unobservable road surface areas.
Solution Approach 2:
The patent replaces direct mechanical/optical detection of road surfaces with a computational approach. Instead of relying on direct image features or edge detection for distant road surfaces, the system substitutes physical measurement with mathematical calculations based on feature point movements and camera geometry, enabling indirect measurement of road surface areas far from the vehicle.
2Measurement precision
If edge observation methods are used to detect road surfaces, then detection accuracy is improved in areas with strong edges, but detection capability deteriorates in road surface areas far from the vehicle where edge observation is difficult
Solution Approach 1:
The patent uses feature points as intermediaries to indirectly detect road surface areas where direct edge observation fails. By tracking feature points and calculating their distances, the system infers road surface boundaries and areas without needing to directly observe edges in distant regions, thus overcoming the limitation of edge-based detection methods.
Solution Approach 2:
The patent substitutes direct edge observation with computational inference. Instead of relying on visual edge detection algorithms that fail in distant areas, the system replaces optical detection with mathematical calculations based on feature point trajectories and camera parameters, enabling road surface detection in areas where edges are not observable.
3Measurement precision
If only feature point-based methods are used to extract road surface areas, then extraction accuracy is improved in areas with sufficient feature points, but extraction capability deteriorates in distant road surface areas where feature points are scarce
Solution Approach 1:
The patent uses feature points as intermediaries to recover road surface area information in distant regions. By calculating distances from feature points to the camera and using geometric relationships, the system infers the extent and position of road surfaces even when feature points are scarce in those distant areas, preventing information loss through indirect reconstruction.
Solution Approach 2:
The patent replaces direct road surface area measurement with computational reconstruction. Instead of relying on direct observation of road surfaces in distant areas, the system substitutes physical measurement with mathematical modeling based on feature point data and camera geometry, recovering road surface area information that would otherwise be lost.
Data Source
AI summary
A vehicle external recognition extracts a feature point from an image including an environment around a vehicle; measures a distance from the vehicle to the feature point based on a movement of the feature point; extracts, as a first road surface area, a local area judged as a road surface based on distance information to the feature point and a position of the feature point in the image from among a plurality of local areas in the image; calculates a multi-dimensional image feature amount including color information for each of the plurality of local areas, and similarity to the first road surface area using the image feature amount with respect to a no-road-surface-extracted area(s) from among the plurality of local areas; extracts a second road surface area from the no-road-surface-extracted area(s) based on the similarity; and recognizes a travelable area using the first and the second road surface areas.