OSOD_Carla

Description : 

Machine learning for autonomous vehicle requires a lot of pictures covering a large amount of situiations. However, some situation called « corner cases » rarely happen in real life while they are very dangerous. OSOD_Carla enables to quickly generate simulated images of « corner cases » whith a great diversity in the situation portrayed.

How does it work : 

OSOD_Carla makes the link between Scenic and Carla to generate simulated images of corner cases.

• Definition of simulation scenarios
• These scenarios are translated into the Scenic language
• Carla simulations are started with parameters comming from Scenic
• Images are captured from Carla simulations

Applications : 

• Creating data base for machine learning applied to autonomous vehicle
• Object detection and recognition
• Generating images of corner cases

Advantages : 

• Multiple paramaters may be set up (road scenario, weather, time of day, etc.)
• Great diversity in little time 
• Works with a docker (not modified by updates of the used softwares)