Contents I. Introduction
With the development of technology, radar sensors have become an indispensable tool in the military, aerospace, meteorology, biomedical, and other fields. Compared to camera sensors radar can undertake a variety of measurement tasks to protect personal information. Radar has advantages in estimating target position and kinematics, and mmWave radar excels in measuring fine motion. mmWave radar is used in angle estimation [1] and localization systems [2], [3], ranging [4], human body recognition [5], [6], gait recognition [7], ball motion studies [8], human activity classification [9], gesture perception and classification [10], posture estimation [11], vibration detection [12], vital sign detection [13], and other fields. Since the COVID-19 epidemic, the focus of attention has gradually returned to physical health. Fitness is one of the most significant ways to build a healthy body. Along with the popularity of the concept of fitness, the fitness industry has developed, and so has the concept of fitness-assisted sensors. The fitness-assisted sensor collects information about the user’s movement, tallies the movement data, and improves the exercise program. The study of repetitive motion counting in fitness-assisted sensors is a relatively new, important, but challenging problem. Among the repetitive motion studies in recent years are contacted sensor-based [14], [15], [16], [17], [18] and smartphone-based [19]. Users must wear the sensors on their bodies when exercising, which increases the discomfort of exercise and may even be a safety hazard. There are also some studies based on video [20], [21], or combining vision and AI [22], [23]. Vision-based methods allow for convenient noncontact measurements, but the equipment is demanding, does not provide privacy, and is susceptible to the background environment. Xiao et al. [24] utilize backscatter from 20 MHz WiFi indoors to identify motion counts within 3 m. There are still system size and privacy issues (network connection is required), and backscattering is susceptible to interference. Fitness assist devices need a contactless solution, that does not violate the user’s privacy and is not easily affected by the environment. The mmWave radar-based solution can solve the above problems. mmWave radar has been applied to classify and identify various types of behaviors [25], [26], [27]. These provide the basis for motion classification and counting. Tiwari and Gupta [28] realize the classification of noncontact motion types based on the CNN network and mmWave radar 2D-FFT Doppler diagram and verify the possibility of mmWave radar as a sports fitness-assisted sensor, but do not conduct a related sports counting research. 2D-FFT does not have azimuth information, has weak spatial information description capability, and cannot perform better-repeated motion counting work. While 3D-FFT can provide higher spatial resolution, accuracy, and detailed estimation of target motion with better tracking performance. Therefore this research aims to validate and establish the possibility of mmWave radar for repetitive motion counting, which is another key feature of fitness-assisted sensors.
