Vanishing Point Detection Using Neural Network Feature Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Determining vanishing points in images for autonomous driving vehicles is challenging due to the difficulty in accurately identifying and indicating these points, which affects the calibration of sensors and the accuracy of neural networks in generating training data.
Innovation Solution
A method is introduced to precisely determine vanishing points in images by using neural networks that represent locations as relative distances to the image center, reducing the problem and solution space, and allowing feature maps from one neural network to be used in another, enhancing the accuracy of sensor calibration and image analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods are used to determine vanishing points in images, then the process is straightforward, but the accuracy of vanishing point detection is poor
Solution Approach 1:
The patent introduces an intermediary neural network that processes the original image to generate a simplified representation containing only relevant features (lines, edges, shapes). This intermediary representation serves as a mediator between the complex original image and the vanishing point detection process, filtering out irrelevant information and highlighting only the features necessary for accurate vanishing point determination.
Solution Approach 2:
The patent segments the image processing task into two distinct stages: first, generating a simplified representation with relevant features using a neural network; second, detecting vanishing points from this simplified representation. This segmentation allows each stage to focus on specific aspects of the problem, improving overall accuracy while managing complexity.
2Productivity
If the problem and solution space is reduced by representing locations as relative distances to image center, then the neural network processing efficiency is improved, but the representation capability is constrained
Solution Approach 1:
The patent changes the parameter representation from absolute coordinates to relative distances from the image center. This parameter transformation reduces the solution space and simplifies the neural network's task, improving processing efficiency. The relative distance representation maintains sufficient information for accurate vanishing point detection while constraining the search space to a more manageable range.
3Productivity
If feature maps from one neural network are used in another neural network, then the processing efficiency is improved, but the accuracy of feature utilization may be compromised
Solution Approach 1:
The patent performs preliminary processing of the image through the first neural network to generate optimized feature maps before these features are utilized by the second neural network for vanishing point detection. This preliminary action prepares the data in an optimal format, ensuring that when features are transferred between networks, they are already refined and ready for accurate utilization, maintaining both efficiency and precision.
Data Source
AI summary
In some implementations, a method is provided. The method includes obtaining an image depicting an environment where an autonomous driving vehicle (ADV) is located. The method also includes determining, using a first neural network, a plurality of line indicators based on the image. The plurality of line indicators represent one or more lanes in the environment. The method further includes determining, using a second neural network, a vanishing point within the image based on the plurality of line segments. The second neural network is communicatively coupled to the first neural network. The plurality of line indicators is determined simultaneously with the vanishing point. The method further includes calibrating one or more sensors based of the autonomous driving vehicle based on the vanishing point.