Sofware components for gesture recognition using 3D-CNN
KOPEČNÝ, L. HAVRÁNEK, Z. BENEŠ, P.
Software components for implementing gestures with convolutional neural network on platform PC with acceleration of graphic card NVIDIA GeForce RTX2080 TI, with possibility of transfer on breadboard, e.g. Jetson-Nano. Software components were developed in programming language Python using the module Keras for neural network processing. There were developed three software components: for creating the training set, for learning phase and for evaluating phase. The first component reads pictures from the beamformer and stores them in form of movies with suffix .avi. The result is a database of movies with gestures. The second component reads those movies and creates the training set of them. The result is a model of neural network accesible in the format .h. The third component reads pictures from beamformer, creates movies of them and evaluates them with trained neural network.
convolutional neural network, gestures, Python, Keras
Vysoké učení technické v Brně, CEITEC VUT Laboratoř pokročilých senzorů, B1.04 Purkyňova 656/123 612 00 Brno