PhD opportunities

Simulation of sensors in point clouds, Photo-realistic rendering of scene by hybrid rendering of rasterization and raytracing in 3D point clouds

Thesis proposal

Area of expertiseReal-time computer science, robotics,systems and control - Paris
Doctoral SchoolSMI - Sciences des Mtiers de l'Ingnieur
SupervisorM. Franois GOULETTE
Co-supervisorM. Jean-Emmanuel DESCHAUD
Research unitMathematics and Systems
ContactDESCHAUD Jean-Emmanuel
Starting dateOctober 1st 2018
KeywordsLiDAR, rendering, mobile mapping, photo-realistic, raytracing, GPU
AbstractThis CIFRE thesis is related to a project that aims to automatically produce 3D databases and a photo-realistic simulation capability of environments for the validation of autonomous vehicles.
The idea is to realistically simulate a set of sensors (camera, lidar, radar) from a 3D scene built entirely automatically from 3D point clouds and images. A stage of classification of the elements of the scene and of distinction of the materials will be the work of another thesis upstream.
OPTIS real-time rendering and virtual reality software are already able to load and display a scatter plot with colorimetric information by maintaining a display rate of more than 20 frames per second for a number of points reaching 300 million in a simulation.
A first step will be to load an even larger dataset, containing in addition the normals and material information at each point allowing re-illumination of the scene. Indeed, with the normal and even basic information on the bidirectional reflectance of the materials, it will be possible to modify the colors that were captured during the acquisition and to make them evolve with the displacement of the light sources or the sensors in the simulation.
A second step will be to adapt the principle of ray tracing on points clouds. To reduce computing time and achieve real-time, one solution will be to use graphics processors (GPUs) instead of conventional processors (CPUs) to take advantage of their massively parallel processing capability.
Finally, OPTIS already has expertise in the field of sensor simulation for vehicles. However, these sensors have not been designed to perform simulations on realistic and noisy environments such as point clouds. Research and development work is needed to adapt existing simulated sensors to these new environments and new simulation techniques.
ProfileGeneral profile:
- Master level 2
- Good interpersonal skills, rigor and autonomy
- Writing and oral presentation skills
- English spoken and written.
- (For foreigners) French spoken and written.

Scientific and technological knowledge used:
- GPU Programming
- Computer Graphics
- 3D data processing
- Image processing

Sotware Developments will be on a PC running Windows or Linux in C ++
FundingConvention CIFRE
PDF document