TAIPEI (Taiwan News) — A team from National Cheng Kung University (NCKU) beat 52 global rivals in a challenge held by the German Association for Pattern Recognition (DAGM) to optimize autonomous driving technologies in a natural environment.
Led by Hsu Chih-chung (許志仲), an assistant professor at the NCKU Department of Statistics, the team comprising four students was recognized in a semantic segmentation challenge launched by the DAGM German Conference on Pattern Recognition (GCPR). The competition seeks to offer solutions to scene recognition difficulties for driverless vehicles.
Existing technologies in scene understanding are mostly applied in urban and structured environments, but fail to address the need for such vehicles to run in unstructured environments, such as the suburbs. Complex scenes in a natural setting can affect the image recognition performance of self-driving cars, cutting the degree of accuracy from 80% to less than 60%.
Incorporating expertise in statistics, electrical engineering, and information management, the NCKU team has managed to work out artificial intelligence modules that can resolve problems with imagery involving bushes, woods, and other natural features through multiple simulations.
The breakthrough serves to boost the safety of autonomous driving while demonstrating Taiwan’s prowess in relevant technologies, according to NCKU.