Synthetic Radar Data Generation via Sensor Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The development of autonomous driving systems faces challenges in generating labeled training data efficiently, particularly for sensor modalities like radar sensors, which are difficult to simulate accurately, leading to high computational costs and time-consuming manual labeling processes.
Innovation Solution
A method that uses a learning algorithm to generate synthetic training data by associating measurements from easier-to-simulate sensors with those from more challenging sensors, such as radar, through a generative adversarial network, allowing for unsupervised learning and reducing the need for manual labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling is used to create training data, then data quality and accuracy are improved, but time consumption and costs increase significantly
Solution Approach 1:
The patent uses a first sensor system to generate synthetic measurement data that copies the characteristics of the second sensor system's measurements. This synthetic data is then used to train the recognition model, replacing the need for manual labeling while maintaining data quality through the copying of realistic sensor characteristics and noise profiles.
Solution Approach 2:
The system performs self-labeling by using the first sensor system's measurements to automatically generate training data for the second sensor system. The recognition model is trained to recognize objects by learning from the correlated measurements of the first sensor system, eliminating the need for external manual annotation services.
2Loss of time
If simulations are used to generate training data, then manual labeling is avoided, but computational effort and resources increase
Solution Approach 1:
The patent introduces a first sensor system as an intermediary that captures real-world measurements which are then used to generate synthetic training data for the second sensor system. This intermediary approach avoids the need for complex physics-based simulations while still producing realistic training data, reducing computational effort compared to full simulations.
Solution Approach 2:
The first sensor system collects and processes measurements in advance to create a dataset that serves as the basis for generating training data for the second sensor system. This preliminary data collection and processing reduces the need for computationally intensive simulations during the actual training phase.
3Reliability
If complex simulations are used to generate realistic sensor data, then data realism is improved, but device complexity and computational resources increase
Solution Approach 1:
The patent merges the measurements from the first sensor system with the training requirements for the second sensor system. By combining real measurements from the first sensor with correlation algorithms, the system generates realistic training data without requiring complex simulations of the second sensor's physical behavior, thus reducing device complexity while maintaining data realism.
Data Source
AI summary
The present disclosure relates to a method for generating training data for a recognition model for recognizing objects in sensor data of a vehicle. First sensor data and second sensor data are input into a learning algorithm. The first sensor data comprise measurements of a first surroundings sensor. The second sensor data comprise a measurements of a second surroundings sensor. A training data generation model is generated, using learning algorithm, that generates measurements of the second surroundings sensor assigned to measurements of the first surroundings sensor. First simulation data are input into the training data generation model. The first simulation data comprise simulated measurements of the first surroundings sensor. Second simulation data are generated as the training data based on the first simulation data using the training data generation model. The second simulation data comprise simulated measurements of the second surroundings sensor.