Neural Network Feature Training for Compact Infrastructure Image Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for aligning images from multiple motor vehicles face challenges in efficiently extracting features suitable for alignment while minimizing data volume to manage limited bandwidth.
Innovation Solution
A method for training an artificial neural network that optimizes feature extraction by incorporating a loss function dependent on both the pose of the feature and the data volume, using a convolutional neural network (CNN) and fully connected network (FCN), with quantization to reduce data size, and defining the loss function to balance pose accuracy and data efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If features are extracted from images for alignment, then alignment suitability is improved, but data volume increases
Solution Approach 1:
The patent changes the parameter of data volume by introducing a loss function that penalizes large feature data volumes. The neural network is trained to minimize both pose error and data volume, resulting in compact feature representations that maintain alignment suitability while reducing data size for transmission.
2Measurement precision
If feature extraction is optimized for alignment, then alignment precision is improved, but bandwidth consumption increases
Solution Approach 1:
The patent implements feedback by using a loss function that provides guidance during training. The loss function combines pose accuracy requirements with data volume penalties, creating a feedback mechanism that steers the neural network to produce features that are both accurate for alignment and compact for transmission.
3Measurement precision
If more detailed features are extracted, then alignment accuracy is improved, but memory requirements increase
Solution Approach 1:
The patent changes the parameter of memory requirements by training the neural network to minimize data volume through the loss function. This results in more efficient feature representations that maintain alignment accuracy while reducing the memory needed for storage and transmission of extracted features.
Data Source
AI summary
A method for training an artificial neural network uses training data that include first image data of a first image and second image data of a second image of an infrastructure. The first image includes a first feature, and the second image includes a second feature corresponding to the first image. The training data include a relative desired translation and a relative desired rotation between the first feature and the second feature. The training includes extracting the first feature from the first image and extracting the second feature from the second image using the artificial neural network. The extracted first feature is represented by first feature data having a first volume of data. The extracted second feature is represented by second feature data having a second volume of data. The training further includes ascertaining a relative translation and a relative rotation between the extracted first feature and the extracted second.