The Joint Urban Remote Sensing Event (JURSE) took place in May 2019 in Vannes, France. It is an international conference with a committee of reviewers focusing on remote sensing and its applications such as building localization in 2D aerial images, slums detection in urban environment, etc.
Berger-Levrault was pleased to present their work on 3D urban object classification. This scientific paper presented results from the Ph.D. thesis of Younes Zegaoui which aims to automatically recognize and localize urban objects (cars, poles, trees, etc.) in LiDAR scanned urban scenes.
In this paper, we argue that for Deep-Learning localization to be functional, it is necessary first to make sure that the simpler task of recognizing isolated objects is validated. Thus, we propose to evaluate the performances of a state-of-the-art 3D neural network as well as additional experiments in order to improving recognition performances. Both of our datasets (train set and test set) are available here.
Berger-Levrault also presented their innovative ideas for automatic urban object monitoring and its usefulness for city managers. In fact, the LiDAR data being geo-referenced, it would be possible to use the network predictions to update existing Geographical Information Systems. Additionally, we could use the same algorithms in an indoor context to perform equipment inventory and tracking.
Link to the conference: http://jurse2019.org/
Link to the scientific paper: http://hal.cirad.fr/lirmm-02087761v1