Object and Direction Classification based on Range-Doppler Map of 79 GHz MIMO Radar Using a Convolutional Neural Network

We propose a convolutional neural network (CNN)which classifies the type and direction of a stationary object from the range-Doppler map acquired by a 79 GHz MIMO radar. The multi-class (21 classes)classification is performed by inputting the single frame range-Doppler map. We measured five kinds of objects oriented in four directions with the MIMO radar equipped with 32 virtual arrays of 4 TX and 8 RX from a distance of 5 m. The CNN with a small number of layers are proposed and range-Doppler maps of the selected virtual arrays are used for input. In the classification of 21 classes, the two convolution layer CNN achieves an accuracy of 92.2% by inputting four virtual arrays in the diagonal direction.