SYNTRA: SYNTHETIC TRAFFIC DATA FOR AI MODEL GENERATION, BENCHMARKING AND OPTIMIZATION
In SYNTRA, NovelSense is developing a Software-as-a-Service (SaaS) solution to generate synthetic traffic datasets to be used in the training, testing and benchmarking of AI-based traffic monitoring systems.
The goal is to make high-quality traffic data available to enable safe autonomous driving. Real data collection is costly and has privacy issues, while existing datasets are limited, unrealistic and often restricted to academic use.
Start date: 02/01/2023
Duration in months: 9
Problem Description
Making high-quality traffic data accessible will allow for safe autonomous driving. Real data acquisition is expensive and has privacy concerns, while existing databases are constrained, irrational, and frequently only useful for academic purposes. To overcome this issue a realistic traffic dataset needs to be created.
Goals
New services
Challenges
The main challenges of the SYNTRA experiment are:- the creation of the SYNTRA Webapp through which customers can acquire customized synthetic traffic datasets for commercial use- the improvement of ABAKUS.AI traffic recognition model through the generated synthetic data
Innovation results
The project will simplify access to high-quality traffic data for training, testing and benchmarking AI models. This helps AI traffic engineers to develop more robust and better characterized models. Municipalities and their citizens benefit from more cost-efficient and accurate real-time traffic monitoring to improve traffic planning and control, or through environment friendly shuttle services which recognize vulnerable traffic participants in complex scenarios.
Business impact
confidential