Neural network based system for human action recognition

1. Emina Petrovic, Univerzitet u Nišu, Mašinski fakultet, Serbia
2. Miloš Simonović, Univerzitet u Nišu, Mašinski fakultet, Serbia
3. Nevena Tomic, Serbia
4. Vlastimir Nikolic, Univerzitet u Nišu, Mašinski fakultet, Serbia
5. Ivan Ćirić, Univerzitet u Nišu, Mašinski fakultet, Serbia

Human action recognition is often used in human-robot interactive applications, such as the visual surveillance, rehabilitation and robotics. Another thing that attracts a lot of attention is a human action recognition based on the trajectory of human tracking. In other words, it is necessary to transform quantitative data of the human movements into qualitative data, so that the numerical characteristics such as speed, direction, etc., of the observed person can be presented as "moving towards something" or "moving away from something" or “standing relative to something”. The research presented in this paper refers to the problem of identifying human action based on information received from the robotic vision system that uses a camera as a sensor, in order to carry out tasks when a robot behaves like a human's assistant. For the purpose of this paper, the classifier based on neural networks was developed in order to classify the human actions.

Ključne rečitestttt :

Tematska oblast: Mechatronics

Datum: 10.03.2017.

13th International Conference on Accomplishments in Mechanical and Industrial Engineering

Datoteka uz rad  (4.45 MB)

Ostali radovi sa konferencije

Pretraži radove