Raw Radar Spike Encoding for Adult–Child Movement Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing radar-based object detection methods, particularly those using Spiking Neural Networks (SNNs), face challenges in accurately recognizing objects due to the need for complex data processing, computational resources, and the loss of meaningful information through data transformations like Fourier transforms, which hinder effective detection and recognition of subtle movements.
Innovation Solution
A method that processes radar signal data without conversion or transformation, utilizing time series information to generate spikes directly from raw radar data, which are then fed into a Spiking Neural Network (SNN) for improved detection and recognition, especially suitable for detecting slight movements in confined spaces.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If Fourier transform or other data conversion methods are applied to radar data, then data processing can be performed, but meaningful information is lost and detection accuracy deteriorates
Solution Approach 1:
Instead of converting radar data to spectrograms or other transformed representations, the patent inverts the conventional approach by directly processing raw radar signal data through spike encoding. This inversion preserves the original signal characteristics and avoids information loss while maintaining processing capability.
Solution Approach 2:
The patent extracts only the essential features needed for detection directly from the raw radar signal through spike encoding, rather than performing comprehensive Fourier transforms. This selective extraction maintains critical movement information while eliminating unnecessary processing steps that cause information loss.
2Measurement precision
If complex preprocessing and data transformation are applied, then feature extraction can be enhanced, but computational effort and processing complexity increase
Solution Approach 1:
The patent replaces complex mechanical signal processing systems (Fourier transforms, spectrogram generation, multiple filtering stages) with a neural spike encoding system. This substitution achieves feature extraction through biological-inspired neural mechanisms that are computationally more efficient while maintaining or improving detection accuracy.
Solution Approach 2:
The patent changes the parameter representation from continuous spectrogram values to discrete neural spike events. This parameter transformation simplifies the data structure and reduces computational complexity while preserving the essential temporal and frequency characteristics needed for accurate feature extraction.
3Extent of automation
If conventional DNN is used to process radar data, then object detection can be performed, but time series information and subtle movements are not effectively captured
Solution Approach 1:
The patent replaces conventional DNN architectures with Spiking Neural Networks (SNNs) that are specifically designed to process temporal sequences of spike events. This substitution enables effective capture of time series information and subtle movements through the inherent temporal processing capabilities of spiking neurons, while maintaining automated detection functionality.
4Use of energy by moving object
If Spiking Neural Networks are used with conventional spectrogram input, then energy efficiency is improved, but recognition accuracy remains at conventional levels
Solution Approach 1:
The patent performs preliminary spike encoding directly on the raw radar signal data before feeding it to the SNN, rather than first creating spectrograms. This preliminary action prepares the data in the optimal format for SNN processing, enabling the network to achieve high recognition accuracy while maintaining energy efficiency through sparse spike-based computation.
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
A computer-implemented method for detecting at least one object or at least one person in a monitored area of a monitoring device, the method comprising the steps of : - receiving (S1) signal data from the monitoring device, - composing (S2) time series data based on the signal data received, - extracting (S3) times series features from the time series data, - generating (S4) spikes from the time series features, - detecting (S5) at least one object or at least one person based on the generated spikes.