Gesture recognition

Course description
Gesture recognition is the scientific and technological field in which learning machines are utilized to understand human motions and interact accordingly. Since «gestures are everywhere », there is a great range of applications, which includes the safeguarding of gestural know-how or even the movement-based interfaces in musical interaction for the Cultural and Creative Industries, the collaborative robotics for the Smart Manufacturing, the interaction between passengers and vehicles for Automotive Industries, etc.. This course presents a generic methodology for gesture recognition that has been validated through both industry-oriented and artistic projects. Topics include: (i) motion capturing, (ii) movement analysis, (iii) feature extraction, (iv) deterministic and stochastic modeling of temporal series, (v) continuous and early recognition. Particular emphasis will be placed on relevant machine learning and computer vision methods for motion tracking. The course will also drawon technological paradigms conceived for real-life situations.

Objectives
This cours aims to provide with all the necessary knowledge for:

  • identifying the appropriate sensor according to the movement scenario and application
  • implementing a motion capture by putting a special focus on vision-based sensors
  • extracting features from the signal based on machine learning methods
  • recognizing isolated and multi-user gestures (using HMMs, GMMs, DTW, etc..) based on training sets with single (one-shot learning) or multiple executions (using C++, Max/MSP etc..)

Academic Instructor
Sotiris Manitsaris|MINES ParisTech