External World Recognition Device Texture Segmentation Parallax Voting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional external environment recognition devices struggle to accurately detect objects with minimal height differences, uniform textures, or low texture differences, such as road boundaries and black vehicles, due to challenges in generating parallax from stereo cameras, leading to incorrect object detection in autonomous driving systems.
Innovation Solution
The device segments external environment information into texture areas and generates parallaxes based on depth distances, using a voting system to enhance object detection by adding additional parallaxes based on texture areas, allowing for improved recognition of various objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional external environment recognition devices use only parallax information from stereo cameras for object detection, then the detection process is simple, but objects with minimal height differences, uniform textures, or low texture differences cannot be accurately detected
Solution Approach 1:
The patent combines multiple detection approaches by merging parallax-based detection with texture area segmentation and additional parallax voting. The object detection unit integrates results from the parallax voting unit (which uses depth distance) and the additional parallax voting unit (which uses texture area information), creating a comprehensive detection system that overcomes the limitations of using only parallax information.
Solution Approach 2:
The patent introduces texture area segmentation as an intermediary step that processes image data to identify distinct texture regions. This segmentation serves as a mediator between the raw stereo camera images and the final object detection, enabling the system to detect objects with uniform textures by first segmenting the image into texture areas before performing parallax voting.
2Reliability
If the vehicle is equipped with multiple cameras to improve object detection performance, then more information is obtained, but the device complexity and cost increase
Solution Approach 1:
The patent changes the processing parameters by introducing multiple voting dimensions (depth distance and texture area) instead of simply adding more cameras. The parallax voting unit votes based on depth distance while the additional parallax voting unit votes based on texture area, creating a multi-parameter voting system that improves reliability without increasing hardware complexity.
3Difficulty of detecting and measuring
If conventional devices rely solely on parallax images for detection, then the processing is straightforward, but boundaries with small level differences and objects with little texture difference are undetectable
Solution Approach 1:
The patent adds another dimension to the detection process by introducing texture area segmentation as a separate voting dimension. Instead of relying solely on the single dimension of parallax (depth distance), the system now performs voting in two dimensions: depth distance (from the parallax voting unit) and texture area (from the additional parallax voting unit). This multi-dimensional approach enables detection of boundaries and objects that are invisible in single-dimension parallax images.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enables accurate detection of objects with low height or uniform textures, enhancing the reliability of autonomous driving systems by correctly identifying potential obstacles and reducing false detections.
Implementation Method 1
The external environment recognition device measures the distance from the camera to an object photographed, using a parallax of an overlapping area of images captured by the two cameras
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An external environment recognition device includes: a texture area segmentation unit that segments external environment information into a plurality of texture areas for individual textures specified from the external environment information acquired from an external environment information acquisition unit; a parallax generation unit that generates a parallax according to a depth distance of an external environment, based on the external environment information; a parallax voting unit that votes a parallax according to a depth distance of an object, the depth distance being specified from the external environment information, in a voting space having one axis representing the depth distance of the external environment; an additional parallax voting unit that votes an additional parallax in the voting space in which the parallax has been voted, the additional parallax being voted based on a texture area; and an object detection unit that detects the object, based on the number of parallax votes given in the voting space, thereby recognizing the external environment.