Vehicle Localization Particle Scoring Using Phase Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing localization techniques for automated vehicles using particle filters suffer from artifacts and aliasing issues, particularly in low-visibility conditions, leading to challenges in accurately scoring particles and maintaining localization.
Innovation Solution
Implementing a phase correlation kernel (SHREK) for particle scoring, which analyzes phase information of sensor and base maps in the frequency domain, combining image data to generate a scoring map that enhances particle identification and scoring accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional scoring techniques based on convolutions of image data in spatial domain are used, then the localization function can operate with standard processing methods, but artifacts and aliasing occur leading to reduced measurement precision
Solution Approach 1:
The patent replaces the traditional spatial domain convolution mechanism with a frequency domain phase correlation mechanism. Instead of convolving image data in the spatial domain using kernel operations, the system transforms the sensed map and base map into the frequency domain using Fast Fourier Transform (FFT), performs phase correlation operations there, and transforms the result back to the spatial domain. This substitution eliminates artifacts and aliasing because phase correlation in the frequency domain is mathematically equivalent to cross-correlation in the spatial domain but avoids the edge effects and artifacts inherent in direct spatial convolution.
Solution Approach 2:
The patent changes the domain parameter from spatial domain to frequency domain for the scoring operation. By transforming the maps into the frequency domain using FFT, performing the correlation operation in this transformed space, and then inverse transforming back, the system achieves superior particle scoring accuracy. The frequency domain representation allows for more robust phase information extraction that is less sensitive to noise and artifacts compared to direct spatial domain methods.
2Measurement precision
If phase correlation operations in frequency domain are implemented, then particle scoring accuracy improves, but computational complexity increases
Solution Approach 1:
The patent substitutes computationally intensive spatial domain convolution with frequency domain phase correlation. The key insight is that convolution in the spatial domain corresponds to multiplication in the frequency domain. By transforming the maps using FFT, performing element-wise multiplication in the frequency domain, and then inverse transforming, the system achieves the same correlation result with better numerical stability and accuracy, while the FFT algorithm itself provides efficient O(N log N) complexity that manages the computational load.
3Adaptability or versatility
If intensity and height scoring techniques are used, then the localization can process standard sensor data, but performance degrades in low-visibility conditions such as rain, snow, and fog
Solution Approach 1:
The patent changes the scoring parameter from intensity and height-based metrics to phase correlation-based metrics. Phase correlation operates on the phase information of the Fourier transforms of the maps, which captures structural and positional relationships more robustly than intensity or height alone. This phase information remains reliable even when intensity values are degraded by rain, snow, or fog, allowing the system to maintain localization reliability in low-visibility conditions where traditional intensity-based scoring fails.
Data Source
AI summary
Embodiments herein include an automated vehicle performing localization functions using particle scoring and particle filters. The automated vehicle performs a phase correlation operation that transforms image data of a sensed map and pre-stored base map from a spatial to frequency domain and combines the transformed maps to generate image data of a correlation map. Estimated location information of particles are compared against sensed data or other data in sensed sub-maps or the correlation map. The automated vehicle may apply an image-convolution scoring map by combining image data of the sensed and base maps in the spatial domain. The autonomy system may calculate entropies for the correlation map and image-convolution scoring map and combines these maps based upon the respective entropies.