DRAMA-2
| Reference number | |
| Coordinator | RISE Research Institutes of Sweden AB - Lindholmen Science Park |
| Funding from Vinnova | SEK 4 745 247 |
| Project duration | November 2020 - December 2022 |
| Status | Completed |
| Venture | Electronics, software and communication - FFI |
| Call | Electronics, Software and Communication - FFI - June 2020 |
| End-of-project report | 2020-02915eng.pdf (pdf, 6899 kB) |
Important results from the project
In this project methodology for creating synthetic data was developed. Prototypes of simulators that can be used to generate synthetic datasets for the development of the object detection and action detection algorithms were developed . Generated synthetic data was investigated and led to the improvements of the algorithm trainings. In the future, simulators are planned to be developed further to be easier to use and to create even more realistic data.
Expected long term effects
Methodology for creating synthetic data was developed. Prototypes of simulators that can be used to generate synthetic datasets for the development of the object detection and action detection algorithms were also developed. Generated synthetic data was investigated and led to the improvements of the algorithm trainings. In the future, simulators are planned to be developed further to be easier to use and to create even more realistic data.
Approach and implementation
Software was selected to be used to create synthetic images; Real images were acquired and pre-processed; Simulator was created to generate images for object detection; Validation and evaluation method was chosen and applied to the created images (this led to several iterations and simulator adjustments); Time series of the body key points that simulate movement in real life were prepared; Series were used to generate datasets for the action detection; Having real and synthetic datasets trainings were performed and results analysed.