Obstacle Segmentation via Road Plane Projection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing obstacle segmentation methods in visual obstacle detection are limited to specific types of obstacles and require multi-frame images, leading to restricted applicability and increased false segmentations.
Innovation Solution
An obstacle segmentation method and apparatus that performs edge detection, line detection, and determines line segment directions on a reference road plane to identify obstacle blocks using a connectivity principle, allowing for segmentation of any type of obstacle in a single frame image with reduced false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature-based segmentation algorithm is used, then segmentation accuracy for specific obstacle types is improved, but adaptability to different obstacle types deteriorates
Solution Approach 1:
The patent transforms the specific feature-based segmentation method into a universal method by using vertical line detection that can identify any obstacle with vertical features regardless of type. The algorithm detects vertical edges and projects them to the road plane, creating a segmentation approach that works for vehicles, pedestrians, and other obstacles without requiring type-specific feature knowledge.
Solution Approach 2:
The patent changes the segmentation parameter from obstacle-specific features (shadows, lamps, color) to a universal geometric parameter (vertical line direction). By detecting vertical edges and their projection directions on the road plane, the method transforms the segmentation criterion into a parameter that applies to all vertical obstacles regardless of their specific characteristics.
2Adaptability or versatility
If movement-based segmentation algorithm is used, then adaptability to different obstacle types is improved, but requirement for multi-frame images increases device complexity
Solution Approach 1:
The patent extracts the essential vertical feature from obstacles and uses it for segmentation without requiring temporal information from multiple frames. By taking out the vertical edge detection and projection to road plane as the core segmentation mechanism, the method eliminates the need for multi-frame input while maintaining adaptability to different obstacle types.
Solution Approach 2:
The patent performs preliminary vertical edge detection and line direction calculation on single-frame images, preparing the segmentation data structure in advance. This preliminary action of detecting vertical lines and projecting them to the road plane enables immediate segmentation without waiting for additional frames, reducing device complexity while maintaining adaptability.
3Productivity
If existing segmentation methods are used, then processing speed is improved, but false segmentation rate increases
Solution Approach 1:
The patent introduces the road plane projection as an intermediary step between edge detection and obstacle identification. By projecting vertical edges to the road plane and analyzing their intersection patterns, the method creates an intermediate representation that reduces false segmentations while maintaining processing efficiency. This intermediary transformation filters out false positives that would otherwise be misidentified as obstacles.
Data Source
AI summary
A barrier partitioning method and device are used for partitioning a barrier area from an image. The method includes: performing edge detecting and partitioning blocks to the image to obtain multiple edge image blocks; performing beeline detection to each edge image block to obtain beelines respectively; for each beeline, determining the line direction of the beeline corresponding to a standard road plane, and determining the edge image block to be a barrier block according to the line direction; forming all the determined barrier blocks to be a barrier area according to a connection principle.