Radar Scatter Plot Transformation for Realistic Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The scarcity and difficulty of obtaining high-quality training data for object recognition from radar signals, due to the specialized knowledge required and sensor-specific distortions, result in residual uncertainty when using synthetic data for real-world applications.
Innovation Solution
A method to generate more realistic simulated radar data by imposing the influence of physical properties from real-world measurements onto simulated data using a Generative Adversarial Network (GAN), transforming point clouds through density distributions to account for sensor-specific distortions and noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If synthetic radar data is used for training, then data availability is improved, but realism and sensor-specific characteristics are worsened
Solution Approach 1:
The patent creates synthetic radar data by generating point clouds that copy the essential structure and characteristics of real radar measurements. By simulating radar reflections, noise patterns, and ghosting effects through computational models, the system produces training data that replicates real sensor behavior without requiring physical measurement campaigns.
Solution Approach 2:
The patent introduces an intermediate processing layer that transforms synthetic point cloud data into a format that better matches real radar measurements. This intermediary transformation applies sensor-specific distortion models, noise characteristics, and signal processing effects to bridge the gap between idealized simulations and actual sensor outputs.
2Measurement precision
If manual annotation of radar data is performed, then training data quality is improved, but time and expertise requirements are worsened
Solution Approach 1:
The patent implements self-service annotation through automated object detection and labeling algorithms that process radar point clouds without human intervention. The system automatically identifies objects, classifies them, and generates ground truth labels by analyzing the spatial distribution and reflection characteristics of points in the synthetic data.
Solution Approach 2:
The patent performs preliminary annotation during the data generation phase by embedding known object positions and characteristics into the synthetic scene models before radar signal simulation. This allows training labels to be automatically associated with generated data without requiring post-processing annotation steps.
3Reliability
If sensor-specific distortions are included in simulation, then training data realism is improved, but simulation complexity is worsened
Solution Approach 1:
The patent segments the complex simulation process into distinct modular components: point cloud generation, radar signal simulation, noise addition, and distortion application. Each module handles a specific aspect of sensor characteristics independently, allowing for systematic implementation and reducing overall system complexity through functional decomposition.
Data Source
Figure 1
Figure 2a~2d
Figure 3a~3d
AI summary
A method (100) for impressing the influence of a physical property (Ia), which is shared by the measurement data (1) obtained by physical measurement and contained in at least one learning scatter plot PL in a domain B, onto simulated measurement data (2) in a simulation scatter plot PA in a domain A, the measurement data (1, 2) in the scatter plots PL and PA in each case representing coordinates, and the method comprising the following steps: • the simulation scatter plot PA is converted into a density distribution pA in the domain A (110); • the density distribution pA is converted by a transformation into a density distribution pB in the domain B (120), wherein this transformation is such that in domain B it is indistinguishable whether a given density distribution p was obtained directly in the domain B as a density distribution pB <sb />of a learning scatter plot PL or as transformation pe of a density distribution pA; • a result scatter plot PB, which is statistically consistent with the density distribution pe is produced in the domain B (130); the result scatter plot PB is assessed as result of the impressing of the influence of the desired property (1a) on the simulated measurement data (2) in the simulation scatter plot PA (140). The invention also relates to: a training method (200); a data set obtained by means of a method (100); a trained AI module and a corresponding data set; a method (300) for identifying objects (5a) and situations (5b); and an associated computer program.