ISAMV

Imaging system for autonomous marine vessels

This project is an exmaple of a system which intend to fuse data from few sensors, in this case a real sense 3D data with data from two thermal cameras in stereo-vision mode, and incorporate machine/deep learning algorithms for obstacle detection & tracking. All the experiments done in a more availble environment and much more work still to be done in testing on real environment, as in imporving the algrothims, their evaluation methods and their results. T

This work shows that even with pretty simple classical approaches and a simple flow design, some start point results can be obtained.

For more details see the report.

Some illusrations from the report:

The setup:

Cameras data (RGB, Depth, Thermal left and right:

Pointclouds :

Fusion :

Deep Learning depth predicted for stereo-thermal images and the built poiintcloud from it :

Obstacales detection & Tracking for Real-Sense RGB & Depth data:

First algorithmical approach:

Second algorithmical approach: