Spatially Invariant Reflective Intensity Volume for Radar Object Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Convolutional neural networks (CNNs) face challenges in object detection using radar signal data due to spatial variation in reflection patterns across the azimuth spectrum, which hinders effective pattern recognition and processing.

Innovation Solution

The method involves generating a spatially invariant reflective intensity volume (RIV) by transforming radar signals into a three-dimensional, spatially invariant range-speed-adjusted-azimuth transform, which is then processed using a trained CNN to detect objects, ensuring consistent spreading functions across different azimuths.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If radar signals are processed using traditional methods preserving spatial variation in azimuth spectrum, then original spatial information is maintained, but CNN pattern recognition effectiveness deteriorates due to inconsistent spreading functions across different azimuths

Engineering Contradiction:

Improveobject detection accuracyVSAvoiddata transformation complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent transforms the azimuth parameter from conventional angular representation to a spatially invariant representation through mathematical transformation. This changes the parameter space of the radar data, making spreading functions consistent across different azimuths and enabling effective CNN pattern recognition while maintaining all necessary spatial information.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent introduces an intermediate transformation step that converts spatially variant radar data into spatially invariant form before CNN processing. This intermediary transformation layer acts as a mediator that reconciles the conflict between preserving spatial information and enabling effective pattern recognition, allowing the CNN to operate on uniformly distributed data.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Productivity

If spatially invariant transformation is applied to radar data, then CNN processing effectiveness is improved through consistent spreading functions, but original azimuthal spatial variation is lost

Engineering Contradiction:

Improveobject detection efficiencyVSAvoidazimuth spatial information

Core Design Contradiction:

ProductivityVSLoss of information

Solution Approach 1:

The transformation to spatially invariant representation changes the parameter space without losing information, as it is a reversible mathematical transformation. The consistent spreading functions enable more efficient CNN processing while the transformation itself preserves all spatial relationships in a different coordinate system, preventing information loss.

Inventive Principle:

Principle #35Parameter changes

3Device complexity

If conventional radar data processing is used without spatial invariance transformation, then processing pipeline remains simple, but CNN recognition capability is hindered by spatially variant spreading functions

Engineering Contradiction:

Improveprocessing pipeline simplicityVSAvoidpattern recognition accuracy

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The patent applies spatially invariant transformation as a preliminary step before CNN processing. This pre-processing action prepares the data in a form that is optimally suited for pattern recognition, ensuring that spreading functions are consistent across all azimuths before the CNN begins its recognition task, thereby improving recognition accuracy without significantly complicating the overall pipeline.

Inventive Principle:

Principle #10Preliminary action

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This approach enables accurate and efficient object detection by converting spatially variant radar data into spatially invariant data suitable for CNN processing, improving the recognition capabilities of CNN systems in automotive radar applications.

Implementation Method 1

reflective radar signals being received by multiple antennas of a radar sensor system

Methodology Applied

Scientific EffectRadar: Radar

Implementation Method 2

a speed ('Doppler') reflective-intensity spectrum includes speeds of the reflection points

Methodology Applied

Scientific EffectDoppler effect: Doppler Effect

Data Source

PatentUS20240210554A1Object detection using convolution neural networks with spatially invariant reflective-intensity data

Publication Date: 2024.06.27 GM GLOBAL TECHNOLOGY OPERATIONS LLC

AI summary

A method that includes obtaining reflective radar signals regarding a scene monitored by a radar sensor system, producing reflective-intensity (RI) data based on those signals, and generating a reflective intensity volume (RIV) based on the reflective-intensity data. The RI data contains spatially invariant spectrums. The method further includes applying a trained convolutional neural network (CNN) on the generated RIV and detecting objects in the scene based, at least in part, the applying of the trained CNN on the generated RIV.