Depth Map Estimation Using Epipolar Feature Map Transformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining depth maps and optical flow in autonomous driving are inaccurate, particularly near the epipole, and require high computational costs.
Innovation Solution
A method involving the generation of transformed feature maps using scale factor candidates, computation of a cost volume, and determination of a disparity map to accurately calculate depth maps and optical flow, utilizing convolutional neural networks and unsupervised machine learning to enhance precision and reduce computational effort.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods are used to determine depth map and optical flow, then the computational process is simpler, but the accuracy is poor especially near the epipole
Solution Approach 1:
The method segments the feature map processing by generating multiple transformed feature maps corresponding to different scale factor candidates. Each transformed feature map represents a different disparity hypothesis, allowing the system to evaluate multiple possibilities independently through cost volume computation, thereby improving accuracy without overwhelming computational complexity
Solution Approach 2:
The invention introduces a new dimension by creating transformed feature maps at different scale factors along the epipolar line. This dimensional expansion in the scale factor space allows the cost volume to capture disparity information more effectively, particularly near the epipole where conventional methods fail, resolving the accuracy-complexity contradiction
2Measurement precision
If the method uses multiple scale factor candidates and cost volume computation, then the accuracy near the epipole is improved, but the computational cost increases
Solution Approach 1:
The method applies local quality by focusing computational resources on regions where they are most needed. The scale factor transformation and cost volume computation are particularly effective near the epipole, providing high accuracy where conventional methods fail, while the structured approach prevents uniform computational overhead across the entire image
Solution Approach 2:
The invention performs preliminary action by pre-computing transformed feature maps for multiple scale factor candidates before the actual disparity determination. This preprocessing step organizes the computational work in advance, allowing the cost volume to efficiently aggregate information without redundant computations during the final disparity calculation phase
Data Source
AI summary
The disclosure relates to a method for determining a depth map and/or an optical flow, comprising providing a first feature map of a first image and a second feature map of a second image, generating a plurality of transformed feature maps from the first feature map and a plurality of scale factor candidates, wherein each of the transformed feature maps is generated by shifting each pixel of the first feature map along an epipolar line by a respective one of the scale factor candidates, computing a cost volume based on the transformed feature maps and the second feature map, and determining a disparity map based on the cost volume, wherein the disparity map specifies the depth map or the optical flow.