DeepSense

The project’s general objective is to design an accurate and robust environment perception system based on the integration of deep learning algorithms with model based probabilistic methods. Due to our experience with ADAS the project will be mainly focused on monitoring and understanding road scenarios: lane detection and tracking, obstacle detection and classification, and traffic sign recognition. The proposed system will be vision based, but will fuse different types of information: camera calibration data, depth data, gyroscope data, car speed, steering angle, throttle position etc.

The proposed framework provides an efficient way to handle uncertainty by combining deep neural networks (network hyper parameters’ uncertainty) with probabilistic inference models (weights uncertainty) and by exchanging information between these two uncertain entities, we will obtain a more precise estimator.

Contract:
PN-III-P1-1.1-TE-2016-0440
NR. 39 ⁄ 2018

Funding:
Ministery of Research and Innovation, CNCS – UEFISCDI