CNN Based Radar Kick Sensor Gesture Recognition Prototype

The concept of kick sensors is aiding users to open or close vehicle doors applying a simple kick gesture using the foot. These sensors have usually been implemented using ultrasound, capacitive sensing, computer vision, and 24GHz Continuous Wave (CW) radar. This paper discusses the algorithm development and implementation of a kick sensor on a 60GHz frequency modulated CW radar platform using deep learning. The goal is to develop a robust yet cost-effective solution for real-time kick gesture recognition. This has been achieved using one transmitting and one receiving antenna, an efficient data compression approach, and a convolutional neural network with a low memory requirement that is capable of achieving 97% accuracy on test data. The final prototype can detect kicks and send control signals to open or close a vehicle’s tailgate at an accuracy level of 88%. Future improvements are discussed as well.