Autonomous Driving Data Pipeline With Frequency-Split Neural Inputs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning systems for autonomous driving face challenges in maximizing signal information from sensor data, as traditional conversion processes can reduce signal fidelity and require new methods for different sensors, leading to inefficiencies in data processing and feature detection.
Innovation Solution
A data pipeline that extracts sensor data into separate components using high-pass, low-pass, and band-pass filters, allowing these components to be pre-processed and fed into different layers of a deep learning network, enhancing signal information retention and computational efficiency for accurate feature detection and vehicle control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional conversion processes are used to make sensor data compatible with deep learning systems, then data format compatibility is achieved, but signal fidelity is reduced
Solution Approach 1:
The patent segments sensor data into multiple frequency components using high-pass, low-pass, and band-pass filters. Each frequency component is then processed separately and fed to different layers of the deep learning network, preserving signal fidelity while achieving format compatibility through targeted processing of specific frequency ranges.
2Adaptability or versatility
If traditional conversion processes are used for different sensors, then data compatibility is maintained, but processing efficiency decreases due to requiring new conversion processes
Solution Approach 1:
The patent implements a universal data pipeline architecture that processes sensor data through a standardized frequency decomposition approach. This multi-functional system can handle different sensor types (cameras, LIDAR, radar) using the same filtering and processing framework, eliminating the need for sensor-specific conversion processes and improving processing efficiency.
3Loss of information
If complete sensor data is provided to all layers of the network, then comprehensive information is available, but computational efficiency decreases
Solution Approach 1:
The patent applies local quality by matching different frequency components of sensor data to specific network layers based on their processing needs. High-pass filtered data (containing edge and feature information) is directed to layers specialized for feature detection, while low-pass filtered data is processed by layers handling global context, optimizing computational efficiency while maintaining information completeness.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image captured using a sensor on a vehicle is received and decomposed into a plurality of component images. Each component image of the plurality of component images is provided as a different input to a different layer of a plurality of layers of an artificial neural network to determine a result. The result of the artificial neural network is used to at least in part autonomously operate the vehicle.