Hand Gesture Recognition for Non-Contact Control of a Technical System
Yevgeniy R. Muratov, Mikhael Nikiforov, Olga V. Melnik · 2020 International Russian Automation Conference (RusAutoCon) · 2020
The article proposes a method and algorithm for determining the hand gesture on a video sequence from only one optical camera. An example of gesture recognition is considered: palm clenching into a fist followed by unclenching of a fist working in real time on Intel Atom and ARM architectures. The method features are low requirements for the resources of a computing device and the possibility of using several gestures for analysis at the same time. At the same time, the method reliably detects gestures against a complex background, including when the camera vibrates, the presence of moving objects in the frame, when the palm covers part of the face, and also when the gesture is performed with a glove. The gesture determination algorithm is based on an analysis of the statistical characteristics of objects in frames. Objects are clusters consisting of closely spaced image areas in several consecutive frames of a video sequence in which motion was detected. The analysis of changes in the characteristics of the cluster reveals events related to the behavior of an object (hand) recorded by the camera. The approach based on the analysis of cluster characteristics allows implementing the embeded algorithm on real-time ARM computers. To increase productivity, the authors propose their own clustering algorithm. And also, an example of applying the implementation algorithm for a gesture of squeezing the palm of a hand into a fist followed by unclenching of a fist.