Stereo Depth Estimation for Parallax-Shadowed Vehicle Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current 3D environmental sensing using stereo camera systems is hindered by uneven illumination caused by parallax, leading to difficulties in determining depth information in shadowed areas.
Innovation Solution
Utilizing a convolutional neural network to analyze geometric information from unevenly illuminated image areas, combining triangulation with geometric evaluation to estimate depth information, and adjusting depth data based on uneven illumination patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If stereo camera systems are used for 3D environmental sensing, then depth information can be determined through triangulation, but uneven illumination and shadows caused by parallax between headlights and cameras make it difficult to determine depth information in certain areas
Solution Approach 1:
The patent segments the image processing task by identifying and separating evenly illuminated areas from unevenly illuminated areas. Different processing methods are applied to each segment: triangulation for evenly illuminated areas and neural network estimation for unevenly illuminated areas, thereby resolving the contradiction by treating different regions differently based on their illumination characteristics
Solution Approach 2:
The patent introduces a neural network as an intermediary component that bridges the gap where traditional triangulation fails. The neural network processes image data from both cameras and generates depth estimates for areas where illumination is uneven, acting as a mediator that compensates for the limitations of the triangulation method in shadowed regions
2Illumination intensity
If headlights are used to generate brightness for image analysis at night, then sufficient illumination is provided, but parallax between the different installation positions of headlights and cameras creates shadows in the image data
Solution Approach 1:
The patent converts the harmful effect of shadows and uneven illumination into a useful signal. By training the neural network to recognize patterns in unevenly illuminated areas, the system learns to interpret shadow patterns as depth information, transforming what was previously a harmful artifact into a beneficial feature for depth estimation
3Measurement precision
If triangulation is used to calculate distance information, then depth can be determined in evenly illuminated areas, but depth information cannot be obtained in unevenly illuminated areas
Solution Approach 1:
The patent creates a universal depth estimation system that can handle both evenly illuminated and unevenly illuminated areas. The system combines triangulation (which works for evenly illuminated areas) with neural network estimation (which works for all areas including shadowed regions), making the overall system multi-functional and capable of operating under various lighting 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
Enables accurate and robust three-dimensional environment perception even in areas where triangulation is not possible, improving separation of foreground and background objects and enhancing depth determination.
Implementation Method 1
the emitter emits electromagnetic radiation, and the first and second receiving sensors receive reflected portions of this radiation
Implementation Method 2
This uneven illumination is caused by parallax due to the spaced-apart arrangement of the receiving sensors
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
The invention relates to a method for determination depth information relating to image information by means of an artificial neural network (2) in a motor vehicle (1), comprising the following steps: - providing at least one emitter (3, 3') and at least one first and one second receiving sensor (4, 5), said first and second receiving sensors (4, 5) being spaced apart from one another (810); - emitting electromagnetic radiation by the emitter (3, 3') (811); - receiving reflected proportions of the electromagnetic radiation emitted by the emitter (3, 3') by the first and second receiving sensors (4, 5) and generating first image information (B1) by the first receiving sensor (4) and second image information (B2) by the second receiving sensor (5) on the basis of the received reflected proportions (S12); - comparing the first and second image information (B1, B2) for determining at least one image area (D1, D2) which is unequally illuminated in the first and second image information and which is produced by the parallax (S13) due to the spaced-apart arrangement of the receiving sensors (4, 5); - evaluating geometric information of the at least one unequally illuminated image area (D1, D2) and estimating depth information by the artificial neural network (2) on the basis of the evaluation of the geometric information of the at least one unequally illuminated image area (S14).