Dynamic Hand Gesture Classification Based on Multistatic Radar Micro-Doppler Signatures Using Convolutional Neural Network

Abstract

We propose a novel convolutional neural network (CNN) for dynamic hand gesture classification based on multistatic radar micro-Doppler signatures. The timefrequency spectrograms of micro-Doppler signatures at all the receiver antennas are adopted as the input to CNN, where data fusion of different receivers is carried out at an adjustable position. The optimal fusion position that achieves the highest classification accuracy is determined by a series of experiments. Experimental results on measured data show that 1) the accuracy of classification using multistatic radar is significantly higher than monostatic radar, and that 2) fusion at the middle of CNN achieves the best classification accuracy.

Automatic Arm Motion Recognition Based on Radar Micro-Doppler Signature Envelopes

“…Since both arms are involved in every gesture and move together either in the same or opposite directions for all suggested motions, then unlike hand motions, there is no information gleaned from angular resolution that would help in improving classifications [21]. Previous techniques for RFbased arm recognition include the work by Sun et al [22] who used five handcrafted MD features and k-NN classifier to recognize seven arm gestures measured by a FMCW radar.…”

Deep-Learning Methods for Hand-Gesture Recognition Using Ultra-Wideband Radar

“…A common approach in radar hand-gesture recognition is to use CNN, which does not require predefined features, but rather, the network self-learns the features from input signals during the training process [18]. The majority of CNN-based hand-gesture recognition methods extract the signature from either: (i) the changes in Doppler over time [18], or from (ii) a snapshot of the overall range-Doppler fingerprint [19]. Both of these signal types are represented in the form of a 2D matrix (monochromatic image) that is further processed by the CNN.…”

Recognition of Dynamic Hand Gesture Based on Mm-Wave Fmcw Radar Micro-Doppler Signatures

“…The emerging deep learning algorithm, such as convolutional neural networks (CNN), has brought a breakthrough in various fields [19], which is regarded as an effective method for dynamic HGR. In [20], a fused CNN architecture is designed for HGR based on [21] proposed a CNN approach for dynamic HGR in driver assistance system. Park et al [22] used the 1-dimension CNN and long short-term memory (LSTM) to learn the waveforms in the 33GHz radar-based HGR system.…”