LiDAR Semantic Segmentation via Two-Step Domain Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing LiDAR semantic segmentation models face challenges in adapting to different domains due to domain differences, requiring extensive training data and time, which is costly and inefficient.
Innovation Solution
A two-step domain adaptation method is employed, where a first LiDAR dataset is converted to align with a second domain's coordinates and resolutions, followed by machine learning to create a semantic segmentation model, and further feature domain adaptation is performed to refine the model for target data, using a deep learning network with encoder layers and occlusion masking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a deep learning model is trained with a large amount of training data to improve segmentation accuracy, then the accuracy is improved, but the time and cost required increase significantly
Solution Approach 1:
The patent applies parameter changes by transforming the first LiDAR dataset from a first domain to a second domain through coordinate conversion and resolution adjustment. This domain adaptation allows the model to train on transformed data that matches the target domain's parameters, reducing the need for extensive training data and time while maintaining high segmentation accuracy in the target domain.
Solution Approach 2:
The patent uses an intermediary approach by introducing a domain adaptation layer that converts the first domain data into the second domain. This intermediary transformation bridge allows the model to leverage existing datasets while adapting them to the target domain's specific requirements, effectively reducing training time and data requirements without sacrificing accuracy.
2Measurement precision
If a deep learning model is trained with a large amount of training data to improve segmentation accuracy, then the accuracy is improved, but the cost increases significantly
Solution Approach 1:
The patent transforms the first LiDAR dataset by adjusting vertical coordinates and resolving horizontal/vertical dimensions to match the second domain's parameters. This parameter transformation enables the model to achieve high segmentation accuracy in the target domain using a smaller quantity of adapted training data, thereby reducing the overall cost of data collection and processing.
Solution Approach 2:
The domain adaptation process acts as an intermediary that bridges the gap between source and target domains. By converting the first domain data into the second domain through coordinate and resolution transformations, the system reduces the quantity of training data needed while maintaining accuracy, thus lowering the cost associated with acquiring and processing extensive training datasets.
3Ease of manufacture
If existing LiDAR datasets are used directly without domain adaptation, then the data collection process is simple, but the model performance deteriorates due to domain differences
Solution Approach 1:
The patent applies parameter changes by converting the vertical coordinates and adjusting the horizontal and vertical resolutions of the first LiDAR dataset to match the second domain's parameters. This transformation enables the system to maintain high segmentation accuracy when using existing datasets, as the transformed data aligns with the target domain's coordinate system and resolution requirements.
Solution Approach 2:
The domain adaptation process serves as an intermediary that reconciles differences between source and target domains. By transforming the first domain data into the second domain through coordinate conversion and resolution adjustment, the system preserves segmentation accuracy while maintaining the simplicity of using existing datasets, as the adaptation layer handles the domain differences automatically.
Data Source
AI summary
A LiDAR semantic segmentation method and a LiDAR-based object perception apparatus. According to an embodiment of the present disclosure, a method for constructing a LiDAR semantic segmentation model through two-step domain adaptation includes converting a first LiDAR data set of a first domain to obtain a second LiDAR data set of a second domain, as a sensor domain adaptation step, performing a machine learning with the second LiDAR data set as training data to obtain a first semantic segmentation model of an artificial intelligence model, and performing a feature domain adaptation for the first semantic segmentation model using target data to obtain a second semantic segmentation model.