Radar Based Gesture Recognition in Embedded Systems

Gesture recognition has played a significant role in human-computer interaction. Millimeter-wave radar sensor is a class of necessary sensor for human gesture recognition in an indoor condition due to its advantages of private protection, low hardware cost. However, for gesture recognition application, it is difficult for us to meet the requirements of high accuracy and real-time simultaneously in embedded system. Unlike other conventional method which increase the cost of hardware of embedded system to improve detection accuracy, in this paper, we proposed a radar-based gesture recognition in low power consumption and low-cost embedded system by using 24 GHz millimeter-wave radar with 1 transmitting antenna and 2 receiving antenna channels. Besides, 2D trajectory information on XY coordination and lightweight deep neural network have been used to distinct ten air-writing numbers. As a result, this system could reconstruct the air-writing symbols and recognize numeric character 0-9 with 95.2% accuracy in real time.