Radar Image Position Estimation With Reliability-Based AI Fallback

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current radar image evaluation methods face challenges in achieving low latency and low energy consumption, particularly in applications requiring efficient processing, such as mobile and vehicle safety, due to computationally intensive signal processing pipelines.

Innovation Solution

A method and apparatus that utilize a trained neural network to estimate object positions in radar images when a reliability criterion is met, falling back to a more resource-intensive radar processing pipeline only when necessary, thereby optimizing energy consumption and processing speed.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If a trained neural network is used to estimate object positions, then energy consumption and processing latency are reduced, but measurement precision may deteriorate compared to traditional radar processing pipelines

Engineering Contradiction:
Improveenergy consumptionVSAvoidposition estimation precision
Core Design Contradiction:
Use of energy by moving objectVSMeasurement precision

Solution Approach 1:

The system dynamically switches between two processing modes: a fast, energy-efficient neural network for normal conditions and a more accurate but resource-intensive traditional radar processing pipeline when reliability criteria are not met. This dynamic adaptation allows the system to optimize energy consumption while maintaining measurement precision when needed.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system changes the processing parameter by selecting different algorithms based on reliability criteria. When the neural network's reliability criterion is satisfied, it uses the neural network; otherwise, it switches to the traditional radar processing pipeline, effectively changing the processing approach to balance energy consumption and precision.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If a traditional radar processing pipeline is used to ensure high measurement precision, then processing latency and energy consumption increase

Engineering Contradiction:
Improveposition estimation precisionVSAvoidprocessing latency
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system dynamically selects the processing pipeline based on real-time reliability assessment. The traditional radar processing pipeline is reserved for cases where the neural network's reliability criterion is not met, minimizing its usage and thereby reducing overall processing latency while maintaining precision when necessary.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

Instead of always using the full traditional radar processing pipeline, the system applies it partially or selectively only when the neural network's reliability criterion is not satisfied. This partial action approach reduces overall processing time and energy consumption while maintaining precision for critical cases.

Inventive Principle:
Principle #16Partial or excessive action

3Productivity

If a trained neural network is used for position estimation, then processing speed increases, but reliability may be insufficient in certain conditions

Engineering Contradiction:
Improveprocessing speedVSAvoidposition estimation reliability
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The system incorporates a reliability criterion assessment that provides feedback on the neural network's output quality. When the reliability criterion is not met, the system triggers a fallback to the traditional radar processing pipeline, ensuring that unreliable neural network results are corrected while maintaining high processing speed for reliable cases.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The system prepares a fallback mechanism in advance by having the traditional radar processing pipeline ready to be activated when the neural network's reliability criterion is not met. This prior cushioning ensures that reliability issues are pre-mitigated without affecting the normal high-speed operation of the neural network.

Inventive Principle:
Principle #11Beforehand cushioning (Prior cushioning)

Data Source

PatentEP4095551A1Method and apparatus to evaluate radar images and radar device
Publication Date: 2022.11.30 INFINEON TECHNOLOGIES AG
  • EP4095551A1 patent drawingFigure 1~2
  • EP4095551A1 patent drawingFigure 3~4
  • EP4095551A1 patent drawingFigure 5

AI summary

A method to evaluate radar images comprises providing at least a first raw radar image and a second raw radar image and determining, whether a reliability criterion is fulfilled. The method further comprises using a first coordinate and a second coordinate output by a trained neural network as an estimate of a position of an object if the reliability criterion is fulfilled, the trained neural network using the first raw radar image and the second raw radar image as an input. Further, the method comprises using a third coordinate and a fourth coordinate output by another radar processing pipeline as the estimate of the position of the object if the reliability criterion is not fulfilled, the radar processing pipeline using the first raw radar image and the second raw radar image as an input.