Image credit: Informs Annual Meeting 2022
Abstract
State of art computer vision and perception models have been developed using publicly available datasets such as Argoverse and ApolloScape. One major limitation of these datasets is the absence of infrastructure information, including lane line details, traffic signs, and intersection information. Such information is necessary and not complementary to eliminate common edge cases. Taking a leap in the future, we introduce a state-of-art synthetically generated dataset with detailed lane and vehicle information for the next generation of self-driving perception and computer vision solutions, named VTrackIt. The main objective of the VTrackIt dataset is thus to enable the development of a new generation of AI/ML solutions that leverage infrastructure information.
This talk was presented at the INFORMS Annual Meeting and was sponsored by the QSR division. Thank you all for attending!
Mayuresh Savargaonkar
Ph.D.
My research interests include, verification and validation of modern systems, electric vehicle charging infrastructure, Li-ion battery prognostics using customized deep learning, and explainable AI.