Deep Learning based visual localization for automotive embedded platform

Date
2024. Jan - 2024. Dec

Visual localization demo

(0:00~0:10) map training mode
(0:10~0:22) localization mode

Purpose of the project

Develop a method for learning environment maps required for autonomous parking
Accurately estimate the current location of the vehicle from a trained environmental map

Contributions

Implemented a deep learning-based visual localization module
Developed a deep learning model structure that operates on an embedded platform for automotive that provides limited operation
Lightweighting and quantizing deep learning models to run in real-time on embedded platforms for automotive applications