Radar Sequence Restoration for Frame-Dropped Gesture Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Radar systems in mobile devices face challenges with limited datasets, random frame dropping, and mixed noise and target motion portions in range-Doppler images, which hinder effective data augmentation and sequence restoration.
Innovation Solution
A computer-implemented method using deep learning techniques, including frequency domain processing and artificial neural networks, to restore sequences by extracting features, performing inverse fast Fourier transformations, and identifying and removing noise portions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Weight of moving object
If radar systems are integrated into mobile devices, then portability and power consumption are improved, but data quality and dataset completeness deteriorate due to frame dropping
Solution Approach 1:
The system performs preliminary actions by capturing multiple frames at higher frame rates than the actual gesture duration requires, and by pre-processing these frames to identify and remove noise portions before the actual classification task. This ensures that even though frames are dropped during storage, the essential gesture information is preserved in the captured data.
Solution Approach 2:
The system creates a synthetic complete sequence by copying and interpolating information from existing frames. The neural network learns to reconstruct missing frames by copying patterns from surrounding frames, effectively generating a complete sequence representation from the fragmented input data.
2Speed
If frame dropping occurs in radar sequences, then processing speed and power consumption are improved, but sequence completeness and information loss worsen
Solution Approach 1:
The system extracts only the essential information from each frame using neural network-based feature extraction, rather than processing every pixel and value. By extracting key gesture features and removing noise portions, the system maintains sequence completeness while reducing the computational burden of processing all captured data.
Solution Approach 2:
The system changes the parameter representation from raw radar frames to extracted feature vectors and frequency domain representations. This transformation allows for more efficient processing by working with compressed, salient features rather than complete frame data, maintaining information integrity while improving processing speed.
3Quantity of substance
If noise portions are included in radar data, then dataset size and training data availability are improved, but measurement precision and classification accuracy worsen
Solution Approach 1:
The system converts the harmful noise portions into beneficial training data by using the neural network to identify and label noise frames. These labeled noise portions are then used to train the system to recognize and reject noise during classification, transforming the problem of noise contamination into an opportunity to improve noise robustness.
Solution Approach 2:
The system applies different processing quality to different portions of the data. Noise portions are identified and removed or corrected, while signal portions are processed with full detail. This local differentiation maintains measurement precision for relevant gesture information while managing the overall dataset composition.
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
Restores full input sequences, enhances data augmentation, and separates target motion from noise, improving the quality and utility of radar data for applications like gesture recognition.
Implementation Method 1
restoring time domain information for the input sequence by performing an inverse fast Fourier transformation on the frequency distribution
Data Source
AI summary
A computer-implemented method for restoring a sequence for a dataset with frame dropping includes receiving an input sequence. A set of features is extracted from the input sequence. A frequency distribution is determined for the input sequence based on the extracted features. Time domain information for the sequence is restored and in turn, data for the input sequence is augmented based on the restored time domain information. Additionally, noise is removed from the input sequence.