Synthetic Vehicle Data Generation for Autonomous Detection Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing autonomous control systems for vehicles require significant resources and time to obtain and label training data for machine-learned models, particularly for object detection and transformation models, as they rely on human operators to annotate images of vehicles, which is inefficient.
Innovation Solution
The use of computer-simulated models of vehicles allows for the generation and labeling of training data automatically, enabling the training of object detection and transformation models without human intervention, using geometric information and multiple perspectives of virtual vehicle models to create large datasets for improved vehicle detection and recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human operators are used to label or annotate training data images, then the accuracy and quality of training data can be ensured, but the time and resources required increase significantly
Solution Approach 1:
The patent uses synthetic images generated from 3D vehicle models as copies of real vehicle appearances. These synthetic images serve as training data without requiring human annotation, as the ground truth information is automatically extracted from the 3D models. This copying approach maintains data quality while eliminating time-consuming manual labeling.
Solution Approach 2:
The system performs self-service by automatically generating synthetic training images and extracting ground truth information from 3D vehicle models without human intervention. The automated pipeline includes rendering images from multiple viewpoints, detecting vehicles, and annotating them using the known 3D model data, thereby eliminating the need for human operators.
2Reliability
If a large amount of diverse training data is obtained to improve model performance, then the accuracy of object detection and transformation models improves, but the resources and time required to collect and label this data increase
Solution Approach 1:
The patent changes the source of training data from real-world captured images to synthetically generated images from 3D models. By adjusting parameters such as vehicle position, orientation, lighting conditions, and background environments in the synthetic generation process, diverse training data is created efficiently without manual collection and labeling efforts.
Solution Approach 2:
The system performs preliminary action by pre-defining 3D vehicle models with accurate geometric information before training begins. These pre-prepared 3D models serve as the foundation for generating unlimited synthetic training images with automatically known ground truth, eliminating the need for time-consuming post-capture annotation work.
3Productivity
If computer simulated models are used to generate training data, then the time and resources for data labeling are reduced, but the complexity of the data generation process increases
Solution Approach 1:
The patent employs a universal 3D vehicle modeling approach that can generate training data for multiple tasks simultaneously. The same 3D vehicle models are used to generate images for object detection, transformation models, and various viewpoints, eliminating the need for separate data collection processes for each task and simplifying the overall workflow.
4Adaptability or versatility
If more viewpoints and perspectives of vehicles are included in training data, then the robustness of the autonomous control system improves, but the amount of data and computational resources required increase
Solution Approach 1:
The patent adds dimensional diversity by generating synthetic images from multiple 3D viewpoints and perspectives. Instead of collecting extensive 2D images from various angles, the system uses 3D vehicle models to render images from any desired viewpoint, efficiently achieving comprehensive coverage with fewer actual images and automatic ground truth extraction.
Data Source
AI summary
A modeling system trains computer models for an autonomous control system using computer simulated models of objects. The objects may be vehicles, and the computer simulated models may be virtual models of vehicles simulated by computer software. Since the vehicle models are computer simulated, various characteristics of the vehicle can be easily obtained by the modeling system. The various types of data may include geometric information of the vehicle, views of the vehicle from different perspectives, and the like. The modeling system can easily generate and label a large amount of training data using the characteristics of the computer simulated vehicles. The modeling system can use the training data to train computer models for the autonomous control system.