Vision-Only Vehicle Training Using Simulated Driving Content
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicles relying solely on vision systems for navigation and safety functions face challenges due to limited data availability and increased detection errors in adverse weather conditions, as they lack the robustness of radar and LIDAR systems.
Innovation Solution
A network-based system generates simulated content using machine learning algorithms trained with labeled data from target vehicles and additional simulated content to enhance vision-only systems, supplementing ground truth labels with generic templates and variations to improve object detection and characterization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If vehicles use vision-only systems for navigation and safety functions, then device complexity is reduced, but detection accuracy deteriorates in adverse weather conditions
Solution Approach 1:
The system performs preliminary actions by collecting and processing ground truth label data from target vehicles, generating content models and simulated content in advance. This pre-processing creates comprehensive training datasets that enable vision systems to maintain high detection accuracy without requiring complex multi-sensor hardware configurations.
Solution Approach 2:
The system creates copies of real-world scenarios through simulated content generation. By replicating adverse weather conditions and various driving scenarios in virtual environments, the vision system can be trained on numerous copies of challenging situations without requiring physical presence in those conditions, thereby maintaining detection accuracy while using simpler hardware.
2Ease of manufacture
If vehicles rely solely on vision systems, then manufacturing cost is reduced, but reliability deteriorates due to limited data availability
Solution Approach 1:
The simulated content generation system serves multiple functions: it generates training data for various weather conditions, creates diverse driving scenarios, produces ground truth labels, and enables transfer learning across different vehicle types. This multi-functional platform improves vision system reliability without requiring expensive specialized hardware for each function.
Solution Approach 2:
The system changes parameters by generating simulated content with varying environmental conditions (weather, lighting, time of day) and object characteristics. By systematically varying these parameters in virtual simulations, the vision system learns to maintain reliable performance across diverse real-world conditions while using cost-effective vision-only hardware.
3Measurement precision
If more training data is collected from target vehicles, then detection accuracy is improved, but data processing time increases
Solution Approach 1:
The system extracts only the essential elements needed for training by generating content models from ground truth label data and synthesizing simulated content with specific focus on relevant objects, scenarios, and environmental conditions. This selective extraction creates efficient training datasets that achieve high detection accuracy without processing unnecessary data, thereby reducing processing time.
Solution Approach 2:
The system performs preliminary processing by pre-generating content models and simulated content datasets before actual vision system deployment. This advance preparation creates ready-to-use training materials that eliminate the need for time-consuming real-time data processing, enabling rapid model training while maintaining high detection accuracy.
Data Source
AI summary
Aspects of the present application correspond to utilization of a combined set of inputs from simulation systems to generate or train machine learned algorithms for utilization in vehicles with vision system-only based processing. Aspects of the present application correspond to utilization of a set of inputs from sensors or sensing systems and simulation systems to create updated training sets for use in machine learning algorithms. The combined set of inputs includes a first set of data corresponding to vision system from a plurality of cameras configured in a vehicle. The combined set of inputs further includes a second set of data corresponding to simulated content systems that generate additional training set data including visual images and data labels to supplement the vision system data.