Homepage Inria website
  • Inria login
  • The Inria's Research Teams produce an annual Activity Report presenting their activities and their results of the year. These reports include the team members, the scientific program, the software developed by the team and the new results of the year. The report also describes the grants, contracts and the activities of dissemination and teaching. Finally, the report gives the list of publications of the year.

  • Legal notice
  • Cookie management
  • Personal data
  • Cookies

Section: New Results

Classification of human actions

It is important for the decomposition of human industrial activities to recognize and classify elementary gestures (a possible decomposition for measuring difficulty is described in section 8.3 or classical methods in industry such as MTM Methods Time Measurement). Due to the temporal nature of the signals, it is necessary to use a type of deep networks that manage this type of data. Recursive networks are therefore used where past observations influence the current prediction. Among recent deep network research, the so-called long-short term memory (LSTM) cells, represented here, seem well adapted. Unlike a simple recursive network where only data from the previous time is used for a new prediction, an LSTM cell can store data over a much longer period of time. With each prediction, the forget gate can decide to authorize the use or forget a previously observed data. We tested our algorithms on a classic benchmark (NTU RGB+D). In order to obtain interesting recognition rates, we showed that it was necessary to use the filters explained in section 7.3 to determinate the number of learning movements. Other less data-intensive methods are to be tested.