Hand Gesture Recognition with FMCW Radar Using Data Reshaping and Machine Learning

Hand gesture recognition, one of the research areas of human computer interaction (HCI) based technology, has become a major focus of attention in the last two decades. Hand gesture recognition, which can be performed with multiple sensors, has found a new sensor especially with the development of radar technology. Hand gesture recognition, which has found its own application areas such as entertainment, security, gesture and posture analysis, is still a very suitable subject to be researched with artificial intelligence. In this study, we used the Dop-net dataset, which contains four different gestures: wave, pinch, swipe, swipe and click from six different individuals. The dataset was generated with frequency modulated continuous wave (FMCW) radar, which has the characteristics of high resolution and easy processing of the output signal. The data is preprocessed with short time Fourier transform (STFT), which is a highly preferred method because it produces two-dimensional output in time and frequency. In this study, it is proposed to reshape the Dop-Net data by downsampling and truncation and to classify them with machine learning algorithms such as support vector machine (SVM) and k-nearest neighbors (k-NN). As a result, the amount of data has been reduced and 93.71% classification accuracy has been obtained with the k-NN algorithm.