Artisense completes ArtiMonoRec research under ELISE grant

Successful Completion of the project ArtiMonoRec funded by the European Network of AI Excellence Centres (elise)

For a period of six month researchers from Artisense were working on the integration of a solely camera based dense reconstruction pipeline to enable accurate perception and 3D mapping. This was made possible due to funding from the European Network of AI Excellence Centres (elise). During the project an AI module, called ArtiMonoRec, was developed that builds on top of the existing Artisense product VINS and delivers highly accurate dense reconstruction of the environment from a moving camera.

ArtiMonoRec in combination with Artisense VINS is able to provide an accurate 3D reconstruction of the environment while running on an embedded AI platform. The real-time output of ArtiMonoRec can be used for online perception to provide e.g. a reliable occupancy model for path planning and obstacle avoidance as well as to build highly accurate 3D maps from image data in a completely automated process. A major differentiator compared to other approaches is that ArtiMonoRec is able to jointly reconstruct the static environment as well as detect and localize moving objects in the scene.

A significant innovation of ArtiMonoRec compared to other neural network based solutions is the proposed training pipeline. The training is defined in a completely semi-supervised manner which does not require LiDAR sensors for data collection. Therefore, the complete system relies on nothing else except the data provided by VINS during training and operation. This makes the system cheaper and much more scalable than other solutions.

The research conducted within this elise grant has achieved all the major milestones that was set out for the project. The team has managed to demonstrate a real-time version of the proposed system running on an embedded platform and has developed it into an option that existing and new customers can consider within their mobile robotic or autonomous driving applications.

ArtiMonoRec is based on the research work MonoRec developed at the Chair of Computer Vision and Artificial Intelligence at TUM in collaboration with Artisense. MonoRec work was presented at CVPR 2021.