Synthetic LIDAR Intensity Correction Using Confidence Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating synthetic LIDAR sensor data are costly, inefficient, and limited by the need for complex material calibration and realistic noise modeling, making it difficult to create accurate virtual vehicle surroundings.
Innovation Solution
A computer-implemented method and system that uses a machine learning algorithm and light beam tracking with precaptured material reflection values to determine intensity values of pixels in a simulated 3D scene, assigning confidence values to calculate corrected intensity values through a weighted mean, allowing for more accurate and efficient generation of synthetic sensor data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex material calibration is performed to model LIDAR intensity in virtual environments, then the accuracy of synthetic LIDAR data is improved, but the cost and time required for data generation increases significantly
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing material properties (albedo, roughness, normal maps) in a lookup table during an offline calibration phase. This allows the online synthetic LIDAR data generation to quickly retrieve pre-computed material parameters without performing complex real-time calculations, thus achieving high accuracy while minimizing generation time.
Solution Approach 2:
The patent uses copying by creating a virtual copy of the real-world environment with digitally reconstructed materials and objects. Instead of physically calibrating each material in the virtual environment, the system copies material properties from real-world measurements and stores them in lookup tables, enabling rapid generation of accurate synthetic LIDAR data without repeated physical calibration processes.
2Reliability
If complex noise and sensor noise profiles are modeled in a model-based manner, then the realism of synthetic data is improved, but the complexity of the system increases
Solution Approach 1:
The patent applies this principle by using simple statistical noise models instead of complex physical noise simulations. Rather than implementing detailed models of sensor noise, multipath effects, and atmospheric interference, the system adds computationally efficient random noise components that approximate real sensor behavior, achieving sufficient realism while keeping the system manageable.
3Manufacturing precision
If complete material property calibration is performed for all objects, then the accuracy of intensity modeling is improved, but the cost and resource requirements increase
Solution Approach 1:
The patent applies local quality by differentiating the level of detail required for different objects in the scene. Critical objects that significantly impact LIDAR intensity measurements (such as vehicles, pedestrians, and road surfaces) receive detailed material calibration with multiple parameters, while less important background objects use simplified material models or default properties from the lookup table, optimizing the balance between accuracy and computational cost.
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
This approach simplifies and cost-effectively generates synthetic LIDAR sensor data with improved accuracy, enabling more precise modeling of virtual vehicle surroundings and reducing the complexity of noise and material property uncertainties.
Implementation Method 1
application of a machine learning algorithm to the distance data, which outputs first intensity values of the pixels
Implementation Method 2
application of a light beam tracking method to the distance data for determining second intensity values of the pixels, using precaptured, in particular calibrated, material reflection values
Data Source
AI summary
A computer-implemented method as well as a system for determining intensity values of pixels of distance data of the pixels generated by a simulation of a 3D scene, including an assignment of a first confidence value to each of the first initial values of the pixels and/or a second confidence value to each of the second intensity values of the pixels, and including a calculation of third, in particular corrected, intensity values of the pixels, using the confidence values assigned to each of the first intensity values and/or second intensity values. The invention also relates to a computer-implemented method for providing a trained machine learning algorithm as well as to a computer program.