Perception & Machine-Learning

Research topics

Robotic systems are generally based on a “Perception → Reasoning → Action” loop. It is therefore natural that the goal of Robotics Lab’s research work in Perception & Machine-Learning is to improve real-time perception, understanding and interpretation of its environment by a robot (in a broad meaning also including “Intelligent” Vehicles). More precisely, scientific contributions in this domain are focused on the following topics:

• Multi-sensor fusion

• Real-time embedded computer vision

• Pattern recognition (categorization or identification)

• Gesture recognition

• Datamining

Regarding the first 2 axes, the aim is to propose new architectures and algorithms for real-time multi-sensor analysis and fusion (vision, lidar, radar, cartographic information, and inertial sensors). The main application domains are design and prototyping of Advanced Driving Assistance Systems (ADAS)  and Intelligent Vehicle (driverless), as well as human gestures recognition, and finally localization and navigation of robots (laser and visual SLAM, both metric and topologic).

Contributions of the 3rd topic (pattern recognition) include real-time detection and recognition in videos of traffic signs (Traffic Sign Recognition, TSR), and of categories of objects such as vehicles and pedestrians, as well as identification of persons (from silhouette or face), and of 3D objects in depth images (3D cameras) or 3D points cloud (laser scanners).

The 4th axis (gesture recognition) focuses on choice and combination of sensors (3D cameras like Kinect or PMD, LeapMotion, inertial sensors, etc…), and on design of movement descriptors and real-time recognition algorithm. Application domains are related either to technical gestures (in particular for human-robot collaboration), or to artistic or handicraft gestures (for  analysis, preservation, and transmission of gestural know-how constituting an Intangible Cultural Heritage), or finally human-machine interactions (gestural control).

Finally, the last topic (datamining) essentially consists in analysis and forecasting of road traffic, and its exploitation: unveiling of spatial and temporal patterns of traffic congestion at the scale of a whole city, large-scale traffic forecasting up to day-long horizon, and optimized cooperative re-routing.



Researchers: Fabien Moutarde, Bogdan Stanciulescu, Sotiris Manitsaris, Amaury Breheret

PhD students: Eva Coupeté, Edgar Hemery, Yannick Jacob, Xiangjun Qian, Olivier Huynh, Manu Alibay, Bruno Ricaud, Florent Taralle, TaoJin Yao.

Ongoing projects



i-Treasures :  Intangible Treasures – Capturing the Intangible Cultural Heritage and Learning the Rare Know-How of Living Human Treasures (FP7 IP)


Helicoid (FP7 FET)

Direct contracts and endowed Chairs:

• Perception axis of “Chaire PSA_Peugeot-Citroën Robotique et Réalité Virtuelle” (RRV)

• Real-time and embedded computer-vision algorithms for Valeo

• Gestural interaction in automotive cockpits for PSA

• Localization and navigation of humanoïd robots in domestic environment (supervision of a CIFRE PhD thesis for Aldebaran)


Past recent projects



Tool : PixelAnnotationTool