Synthetic Multi-Return Lidar Data from Single Return Input
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in creating accurate high-definition (HD) maps lies in the costly and equipment-dependent collection of multi-return Lidar data, which is essential for precise mapping but requires specialized and expensive hardware.
Innovation Solution
A system utilizing a generative network and discriminator network, trained adversarially, to generate synthetic multi-return Lidar data from single return Lidar data, allowing for the creation of HD maps without the need for expensive multi-return Lidar equipment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multi-return Lidar equipment is used to collect data, then mapping precision and ability to capture hidden objects are improved, but equipment cost and complexity increase
Solution Approach 1:
The patent uses a generative adversarial network to create synthetic multi-return Lidar data that copies the characteristics of real multi-return data. The generator network learns the distribution patterns of genuine multi-return Lidar returns and synthesizes new realistic returns, enabling the system to achieve multi-return mapping precision using only single-return Lidar equipment.
Solution Approach 2:
The patent replaces expensive multi-return Lidar equipment with inexpensive single-return Lidar sensors. By using the generative network to synthesize additional returns from single-return data, the system achieves the functionality of expensive equipment using cheap, readily available sensors that can be deployed more widely.
2Productivity
If multi-return Lidar systems are deployed, then data collection efficiency and depth information capture are improved, but equipment cost increases
Solution Approach 1:
The generative network synthesizes additional Lidar returns by learning the statistical patterns and spatial relationships from training data. It creates realistic second and third returns from single-return inputs, effectively copying the information that would be captured by expensive multi-return hardware while using affordable single-return sensors.
Solution Approach 2:
The patent transforms single-return Lidar data into multi-return data by applying learned parameters from the generative network. The system changes the data structure from single return to multiple returns by adding synthesized returns with appropriate intensity, range, and spatial characteristics based on the trained model's understanding of Lidar return patterns.
3Measurement precision
If physical collection by specialized equipment is used, then data accuracy is improved, but cost and accessibility worsen
Solution Approach 1:
The system creates accurate synthetic multi-return data by copying the essential characteristics of real multi-return Lidar returns. The generative network learns the distribution, intensity patterns, and spatial relationships of genuine returns and reproduces them with high fidelity, achieving data accuracy comparable to real multi-return data while using accessible single-return equipment.
Data Source
AI summary
System and methods for creating multi-return map data using single return Lidar data. The systems and methods use a long short-term memory (LSTM) model in combination with a Generative Adversarial Network (GAN) model. The systems and method use a single (1st) return of Lidar at a time stamp and create multiple unseen samples of 2nd and 3rd returns. The LSTM model is used to create a sequential calibration based on incidence angle to choose the optimized 2nd and 3rd return at the same instance of the time stamp. This creates a localized model of three returns from a single return of Lidar and thus provides additional data to generate an HD map.