Postdoct position at LASMEA

Postdoct position: Development of 3D reconstruction algorithms for ultra-fast smart camera in SoC environment

The purpose of this post-doc (proposed in VIRAGO project) is to use a smart stereo camera (designed at LASMEA) based on the reconfigurable logic and CMOS imagers. The main objective is to use a methodology matching the architecture and vision algorithms, to develop an optimal system of high speed embedded vision.
In fact, in conventional approaches, the use of cameras can be summarized by image storage and an "offline" processing. However these systems are expensive and completely inconsistent with the "real time" control of robots. Also, for direct use in the control loop of robot, it is necessary to have real time "sensor" information.
The approach proposed here is to process the large amount of information directly into the camera to eliminate the "bottleneck" between the sensor and control system. The second point is to use sensors using "Rolling Shutter" CMOS technology. The project VIRAGO is built on a major innovation (winner of a European Conference on Computer Vision 2006) which involves the exploitation of algorithm using this technology. In the "rolling shutter" mode, all pixels of the imager are not simultaneously exposed to light but sequentially line by line. When the relative speed between camera and scene becomes very large, deformations or distortions appear in the image. These distortions are not taken into account in the classical perspective projection which considers the observation of a rigid world. This creates unusable images from cameras rolling shutter during pose estimation, and hence for three-dimensional reconstruction. This model allows calculating the accurate placement of the object but also operates the distortions induced by this phenomenon for measuring the instantaneous speed of 3D rigid objects.
In this case, the vision system is planned to have the low-level processing and technical computing compensating defects closer to the retina so as to provide not an image (which must be transferred to PC for processing) but a direct geometrical information (position, orientation, speed). This should provide such information at the rates of 1000 Hz, consistent with the rapid control of robots, and details that are usually less than a few Hertz frequency.

Contacts :
François Berry - Francois.Berry@lasmea.univ-bpclermont.fr – Tel : 04 73 40 72 52
Omar AIT AIDER - omar.ait-aider@lasmea.univ-bpclermont.fr – Tel : 04 73 40 55 67

Pre-requisite:
Proficiency in English(spoken and written) is obligatory.
The candidate should be experience in :
• Hardware Programing (VHDL)
• Electronic System concepts
• System Programing (in C)

Knowledge of following will be plus :
• Artificial vision
• Geometry and calibration

Bibliography
Simultaneous object pose and velocity computation using a single view from a rolling shutter camera
O. Ait-Aider, N. Andreff, J.M. Lavest and P. Martinet
9th European Conference on Computer Vision (ECCV’06), Graz, Austria, May 7-13, 2006.

Structure and Kinematics Triangulation with a Rolling Shutter Stereo Rig
O. Ait-Aider and F. Berry
International Conference on Computer Vision (ICCV'09), Kyoto, Japan, October 2009.
Embedded active vision system based on an FPGA architecture
P. Chalimbaud, F. Berry
EURASIP Journal on Embedded Systems, 2007
Towards an embedded Visuo-Inertial Smart Sensor
P. Chalimbaud, F. Marmoiton and F. Berry
International Journal in Robotic Research 2007