Super-Resolution Training for Obfuscated Traffic Sign Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing super resolution techniques fail to enhance image data resolution sufficiently for applications like crowdsourced mapping and cannot compensate for occluded objects, particularly in low-resolution images with obfuscated traffic signs.
Innovation Solution
A super resolution system utilizing obfuscated image data models and focused loss models to train neural networks, which increase the resolution of low-resolution images by minimizing total, focused mean squared error, perceptual, and variance losses, with weighted factors emphasizing the obfuscated object area.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing super resolution techniques are used to enhance image data, then resolution is improved to some extent, but the resolution improvement is insufficient for applications like crowdsourced mapping and cannot compensate for occluded objects
Solution Approach 1:
The system performs preliminary actions by training the neural network with obfuscated low-resolution data before actual object detection. The training phase pre-processes and adapts the model to handle obfuscated inputs, enabling it to reconstruct occluded objects effectively during deployment without requiring real-time complex processing
Solution Approach 2:
The system changes parameters by transforming low-resolution images into high-resolution representations through neural network processing. The model learns to map low-resolution input space to high-resolution output space, effectively changing the resolution parameter while maintaining object integrity even when obfuscated
2Device complexity
If cameras capture low-resolution image data, then device complexity and data transmission requirements are reduced, but object detection accuracy deteriorates especially for obfuscated traffic signs
Solution Approach 1:
The system replaces mechanical solutions (higher resolution cameras, oversampling) with an intelligent software-based neural network approach. Instead of increasing hardware complexity to capture better images, the system uses a trained model to computationally reconstruct high-resolution images from low-resolution inputs, substituting mechanical enhancement with algorithmic processing
3Reliability
If cameras oversample geographical areas to ensure sufficient image data, then object detection coverage is improved, but data campaign duration and communications bandwidth requirements increase
Solution Approach 1:
The system changes the resolution parameter through neural network processing, allowing standard-resolution cameras to produce detection-quality images via super-resolution reconstruction. This eliminates the need for oversampling, as each captured image is sufficiently enhanced for accurate object detection, reducing both data volume and campaign duration
Data Source
AI summary
A super resolution system that increases a resolution of image data captured by one or more cameras include one or more controllers including one or more super resolution neural networks that include at least one of an obfuscated image data model and a focused loss model. The one or more super resolution neural networks receive paired training data during a training phase, where the paired training data is representative of the image data captured by the one or more cameras representing a surrounding environment and includes low-resolution image data and high-resolution image data. The obfuscated low-resolution image data and the high-resolution image data both represent identical images, and the obfuscated low-resolution image data includes an object of interest located in the surrounding environment that is obfuscated based on an obfuscation technique.