Online hand position detection and classification system using multiple classification algorithms

Hand position recognition is very significant for human-computer interaction. Different kinds of devices and technologies can be used data acquisition; each has its specification accuracy, one these Kinect V2 sensor. A three-dimensional location the skeleton joints taken from device to create three types data, first joint raw second angles between joints, third combined both types. These are train four classifiers, which support vector machines, random forest, k nearest neighbors, multilayer perceptron. The experiments done on datasets 30,480 frames 127 volunteers with saved trained models predict classify eight positions hand in a real-time system. results show that our proposed approach performs well highly efficient accuracy reaching up 99.07% some cases an average time spent checking frame by sequentially short period, cases, it reaches 0.59*10-3 seconds. This system many applications such as controlling robots or devices, comparing physical exercises, even monitoring elderly patients, more.