Stereo Camera Parallax Detection with Distance-Specific Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing three-dimensional object detection systems using stereo cameras face challenges in accurately detecting objects in complex images, particularly in distinguishing obstacles based on distance distribution, especially when objects are far away or have varying heights.
Innovation Solution
A stereo camera apparatus equipped with two cameras that generate a parallax image by calculating the difference in subject position between images from different viewpoints, which is then processed to create partial regions and distance histograms, allowing for the use of distance-specific thresholds to accurately discriminate three-dimensional objects from the environment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single threshold is used for detecting three-dimensional objects across all distances, then the detection process is simple, but detection accuracy deteriorates for objects at different distances
Solution Approach 1:
The patent divides the captured image into multiple partial regions based on distance from the vehicle, with each region corresponding to a specific distance range. Different thresholds are then applied to each partial region, allowing optimized detection parameters for objects at different distances. This segmentation resolves the contradiction by maintaining simple overall process structure while enabling distance-specific detection accuracy through regional differentiation.
2Productivity
If distance distribution information is calculated across the entire captured image, then comprehensive object detection is achieved, but false recognition of road or flat surfaces increases
Solution Approach 1:
The patent segments the captured image into multiple partial regions based on distance, processing each region independently with appropriate thresholds. This prevents false recognition of road or flat surfaces by applying distance-aware detection criteria to each region, while maintaining comprehensive coverage through systematic processing of all regions.
Solution Approach 2:
The patent applies different detection thresholds and criteria to different partial regions based on their distance characteristics. Each region receives localized detection optimization, with closer regions using different thresholds than farther regions. This local quality approach reduces false recognition while maintaining comprehensive detection coverage.
3Adaptability or versatility
If three-dimensional objects are detected using traditional stereo camera methods, then basic object detection is achieved, but detection capability deteriorates for objects with varying heights or at far distances
Solution Approach 1:
The patent segments the detection task by creating multiple partial regions based on distance, with each region processed independently using optimized thresholds. This segmentation enables the system to maintain basic detection capability while significantly improving precision for objects with varying heights and at different distances, as each region can be detected using parameters optimized for its specific distance range.
Solution Approach 2:
The patent changes detection parameters (thresholds) based on the distance of objects being detected. By adjusting thresholds according to the distance-specific characteristics of each partial region, the system adapts its detection capability to handle objects with varying heights and at different distances more effectively, resolving the contradiction between basic capability and precision for varying conditions.
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 enhances the ability to detect three-dimensional objects across a range of distances, reducing false recognition of road or flat surfaces and improving detection accuracy from close to far distances, while ensuring that only objects of interest are identified.
Implementation Method 1
a parallax image is generated by calculating a difference in a position of a subject in the first image and a position of the subject in the second image
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A three-dimensional object detection apparatus (12) includes an input interface (15) and a controller (16). The input interface (12) accepts input of first and second images having mutual parallax in a first direction. The controller (16) generates a parallax image that uses a pixel value to represent a difference in the first direction between a position of a subject in the first image and a position of the subject in the second image and divides at least a portion of the parallax image into a plurality of partial regions extending along a second direction intersecting the first direction. For each partial region, the controller (16) segments a distance in the direction towards the subject into a plurality of distance intervals, calculates, from the pixel value, the distance interval to which each pixel of the partial region belongs; and calculates, on the basis of a pixel count of pixels belonging to the distance interval, an evaluation value corresponding to a length in the second direction of a three-dimensional object belonging to the distance interval. The controller determines whether a three-dimensional object is present in each distance interval on the basis of the evaluation value. The controller (16) uses a different threshold for each distance interval to determine that a three-dimensional object is present when the evaluation value is greater than the threshold.