Low Complexity Radar Gesture Recognition Using Synthetic Training Data

Abstract

Developments in radio detection and ranging (radar) technology have made hand gesture recognition feasible. In heat map-based gesture recognition, feature images have a large size and require complex neural networks to extract information. Machine learning methods typically require large amounts of data and collecting hand gestures with radar is time- and energy-consuming. Therefore, a low computational complexity algorithm for hand gesture recognition based on a frequency-modulated continuous-wave (FMCW) radar and a synthetic hand gesture feature generator are proposed. In the low computational complexity algorithm, two-dimensional Fast Fourier Transform is implemented on the radar raw data to generate a range-Doppler matrix. After that, background modelling is applied to separate the dynamic object and the static background. Then a bin with the highest magnitude in the range-Doppler matrix is selected to locate the target and obtain its range and velocity. The bins at this location along the dimension of the antenna can be utilised to calculate the angle of the target using Fourier beam steering. In the synthetic generator, the Blender software is used to generate different hand gestures and trajectories and then the range, velocity and angle of targets are extracted directly from the trajectory. The experimental results demonstrate that the average recognition accuracy of the model on the test set can reach 89.13% when the synthetic data are used as the training set and the real data are used as the test set. This indicates that the generation of synthetic data can make a meaningful contribution in the pre-training phase.

Adaptive Trajectories’ Constant False Alarm Rate Mirror Filters and Elevation Angle Evaluation for Multiple-Input Multiple-Output Radar-Based Hand Gesture Recognition

“…There are several existing radar-based gesture recognition algorithms. They are based on different hardware architectures, including the continuous wave (CW) radar [10][11][12], the frequency-modulated continuous wave (FMCW) radar [7,8,[13][14][15][16][17][18][19][20], and the multipleinput multiple-output (MIMO) radar [8,21,22]. The CW radar is simple to design and has excellent Doppler sensitivity, but it lacks the resolution of range.…”

“…They used the NN as the feature extractor. Zhao et al [16] proposed a gesture detection system using the FMCW radar. They implemented Blender 4.0 to generate different hand gestures and trajectories.…”

“…There are many classical classification algorithms in machine learning, including neural networks (NNs), support vector machines (SVMs), and random forests. The SVM was adopted for hand gesture recognition in [16]. Based on the random forest, study [9] aimed to determine five different emotions by calculating the relative positions of joints in the arms captured from the Kinect sensor and achieved an accuracy of 86.8%.…”

Dynamic Gesture Recognition Based on FMCW Millimeter Wave Radar: Review of Methodologies and Results

“…Similar to the analysis of feature fusion methods in Section 4.2, due to the different experimental conditions of classification algorithms, we cannot judge the advantages and disadvantages of different classification methods by directly comparing the results of different studies, but we can find the common values or rules from the statistics. The widely used machine learning methods in gesture recognition are support vector machine (SVM) [27,73], K-nearest neighbor method (KNN) [66], and hidden Markov model (HMM) [74]. These methods are easy to implement but lack robustness and computational efficiency.…”

“…Additional studies have enhanced the DTW algorithm for gesture recognition by adding path constraints and refining the matching procedure for gesture recognition [81]. According to the data in statistics [27,44,66,73,75], it can be found that machine learning algorithms are more suitable for problems with simple gestures, small sample categories, and quantity in gesture recognition, and deep learning algorithms tend to have better performance in gesture recognition problems with complex gestures and a larger number of samples.…”