Virtual Sensor Data Generation Using AI Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing solutions for generating output data of virtual sensors do not accurately account for inaccuracies in real sensor measurements, material properties of objects, angles of incidence, interactions with mirrors, transparent objects, and multiple object paths, leading to suboptimal performance in applications like virtual/augmented reality and SLAM.
Innovation Solution
A method and device that simulate a 3D virtual space corresponding to a real space, define environment conditions and object parameters, and generate output data for a virtual sensor by incorporating inaccuracies from real sensor outputs, using a trained AI model to process simulated data and account for various environmental and object interactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If virtual sensor data is generated using basic virtual space simulation, then generation speed is improved, but measurement precision deteriorates because real sensor inaccuracies are not accounted for
Solution Approach 1:
The patent creates a copy of the real sensor's characteristics by training a neural network model on actual sensor data including its inaccuracies and noise patterns. This virtual copy reproduces the real sensor's behavior, allowing generation of realistic sensor data without physical sensors, thus maintaining both speed and precision.
Solution Approach 2:
The patent transforms the sensor data generation process by changing from deterministic geometric simulation to probabilistic generation using neural networks. The model learns parameter distributions from real data, enabling generation of varied yet realistic sensor outputs that capture measurement uncertainties while maintaining generation efficiency.
2Measurement precision
If comprehensive environmental factors are included in simulation, then measurement precision is improved, but device complexity worsens
Solution Approach 1:
The patent introduces a neural network model as an intermediary that encapsulates complex environmental interactions. Instead of directly simulating light propagation, material properties, and sensor physics, the trained model serves as a mediator that generates realistic sensor data by learning these complex relationships from training data, simplifying the overall system architecture.
Solution Approach 2:
The patent performs preliminary training of the neural network model offline using comprehensive environmental data and real sensor measurements. This preliminary action pre-computes the complex relationships between environmental factors and sensor outputs, allowing the deployed system to generate accurate data without real-time complex simulations, thus reducing operational complexity.
3Reliability
If real sensor inaccuracies are incorporated into virtual sensor generation, then reliability is improved, but manufacturing precision worsens due to difficulty in replicating real sensor characteristics
Solution Approach 1:
The patent employs feedback mechanisms during neural network training by comparing generated virtual sensor data against actual real sensor measurements. The model continuously adjusts its parameters based on the difference between simulated and real outputs, ensuring that the virtual sensor accurately replicates real sensor characteristics including their specific inaccuracies and noise patterns.
Solution Approach 2:
The patent uses excessive action by training the neural network on abundant real sensor data from multiple sources and conditions. This over-training approach ensures the model captures all nuances of real sensor behavior, including rare error patterns, thereby improving the reliability of generated data even though it requires more training data and computational resources.
Data Source
AI summary
The disclosure relates to robotics, computer vision, scanning of three-dimensional (3D) objects, navigation, and, in particular, to a method and device for generating output data of a virtual sensor. A technical result is to increase the accuracy of matching the generated output data of the virtual sensor with the output data of the particular real sensor. A method for generating output data of the virtual sensor is provided. The method includes simulating a three-dimensional (3D) virtual space corresponding to a real space, objects having object parameters and the virtual sensor in the simulated 3D virtual space, defining environment conditions of the simulated objects and relative position of the simulated objects in the simulated 3D virtual space and generating the output data of the virtual sensor based on the simulated virtual sensor.