Traffic Simulation Data Augmentation via Point Cloud Obstacle Insertion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data augmentation methods in traffic simulation, such as scaling or rotating images, are insufficient in generating diverse real data due to limited labeling data, hindering the creation of a large amount of realistic data.
Innovation Solution
A data augmentation method that involves acquiring a point cloud of obstacles using a radar, removing original obstacles to fill voids with the surrounding environment, and adjusting new obstacles based on labeling data including position, orientation, and type to generate more diverse layout data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If scaling or rotating a frame of image is used for data augmentation, then the amount of real data increases slightly, but the diversity of real data remains insufficient
Solution Approach 1:
The patent creates synthetic obstacle data by copying and pasting obstacle templates into the background image. Multiple copies of the same obstacle template can be placed at different positions, and different templates can be used to create variety. This approach generates a large quantity of augmented data while maintaining diversity through template selection and placement variation.
Solution Approach 2:
The patent changes multiple parameters of obstacles including position coordinates, size dimensions, orientation angles, and obstacle types. By systematically varying these parameters, the method generates diverse synthetic data that differs significantly from simple geometric transformations, thereby improving both quantity and diversity of augmented real data.
2Ease of manufacture
If simple geometric transformations are applied to existing images, then data augmentation is achieved with minimal modification, but the amount of new diverse data generated is limited
Solution Approach 1:
The patent employs copying of obstacle templates into the background, which is computationally efficient and easy to implement. Multiple obstacles can be generated by repeatedly copying templates and adjusting their parameters, thereby achieving easy data augmentation while generating a large quantity of new diverse data through systematic parameter variation.
Solution Approach 2:
The patent uses pre-prepared obstacle templates that contain obstacle information. These templates are prepared in advance and can be quickly instantiated multiple times in the background image. This preliminary preparation enables efficient generation of large amounts of augmented data without requiring complex processing during the augmentation process.
3Quantity of substance
If obstacle templates are copied and pasted into the background, then the amount of real data increases, but the labeling data requires comprehensive adjustment for different scenarios
Solution Approach 1:
The patent systematically adjusts labeling data parameters including position coordinates, size dimensions, orientation angles, and obstacle types to match different scenarios. By changing these parameters comprehensively, the method generates diverse synthetic data with appropriate labels, thereby increasing the amount of real data while managing the complexity through parameterized adjustment rather than manual labeling.
Solution Approach 2:
The patent dynamically adjusts obstacle parameters such as position, size, and orientation based on the background image characteristics and desired scenario. This dynamic adjustment allows the same template to be adapted to various contexts, increasing data quantity while reducing the need for separate templates for each scenario, thereby managing labeling complexity.
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 method increases the amount and diversity of real data in traffic simulation by arranging and adjusting new obstacles in a realistic background, enhancing the simulation's data richness.
Implementation Method 1
the acquisition vehicle obtains the point cloud by scanning the plurality of original obstacles around the acquisition vehicle with a radar
Data Source
Figure 1~2
Figure 3
AI summary
A data augmentation method, device are provided according to embodiments of the present application. The method includes: acquiring a point cloud of a frame, the point cloud comprising a plurality of original obstacles; obtaining a plurality of position voids by removing the original obstacles from the point cloud, and filling the position voids to obtain a real background of the point cloud; arranging a plurality of new obstacles labeled by labeling data, in the real background of the point cloud; and adjusting the new obstacles based on the labeling data of the new obstacles to obtain layout data of the new obstacles. The amount of real data is increased, and a diversity of the real data is improved.