Can we use modern deep learning methods, to avoid the use of expensive hardware and computationally-expensive signal processing methods for object detection?
One can train models on mini-Doppler maps, collected via software defined radios. One crucial idea is to try and operate in less crowded frequency ranges like 433 MHz, allowing us to use inexpensive off-the shelf hardware. This would allow Deep Learning methods to be combined with inexpensive, low-frequency radars to achieve high accuracy in real-time on various useful tasks.